1. science
2. mathematics
3. algebra

PhDs in Logic
VII
TU Wien,
May 14–16, 2015
Support by the following organisations is gratefully acknowledged:
• Vienna Center for Logic and Algorithms (http://vcla.at/)
• Doctoral Program Logical Methods in Computer Science (http://logic-cs.at/phd/)
Scientific committee:
Agata Ciabattoni (Vienna University of Technology)
Jan van Eijck (CWI, Amsterdam)
Chris Fermüller (Vienna University of Technology)
Nina Gierasimczuk (University of Amsterdam)
Martin Goldstern (Vienna University of Technology)
Thomas Icard (Stanford University)
Eric Pacuit (University of Maryland)
Jakub Szymanik (University of Amsterdam)
Anna Zamansky (University of Haifa)
Advisory board:
Helmut Veith (Vienna University of Technology)
Stefan Szeider (Vienna University of Technology)
Organizing committee:
Ronald de Haan (Vienna University of Technology)
Martin Kronegger (Vienna University of Technology)
Contents
Program
5
Invited Talks
8
Querying Ontology Knowledge Bases using Datalog
Thomas Eiter
9
Towards a Logic-Based Framework for Analyzing Stream Reasoning 10
Thomas Eiter
Answer Set Solving in a nutshell
Torsten Schaub
11
Complexity in Logic: off the beaten track towards pastures new
Uli Sattler
12
New directions in typelogical semantics
Michael Moortgat
13
Cut-elimination in generalisations of the sequent calculus
Revantha Ramanayake
14
Winners of the VCLA International Student Awards
16
Boolean Circuit Optimization
Sophie Spirkl
17
Deciding Properties of Automatic Sequences
Luke Schaeffer
18
SAT Sampling: From Theory to Practice
Kuldeep S. Meel
19
New Complexity Bounds for Evaluating CRPQs with Path Comparisons
20
Pablo Munoz
Contributed Talks
21
Intertranslatability of Abstract Argumentation Frameworks
Sylwia Polberg
22
2
Combining Different Forms of Defeasible Reasoning in Abstract Argumentation: Integrating Mechanisms from Adaptive Logics
26
Jesse Heyninck
An Abstract Algebraic Logic view of propositional-attitude aggregation theory
30
María Esteban and Zhiguang Zhao
Basic model theory of modal languages with propositional constants 33
Matteo Pascucci
Modal Logics for Qualitative Spatial Reasoning
Przemysław Andrzej Wałęga
36
Algebra-based canonical formulas for superintuitionistic logics
Julia Ilin
40
Towards Reconciling SPARQL and Certain Answers
Shqiponja Ahmetaj and Wolfgang Fischl
43
Formalising the SECD machine with Nominal Isabelle
Gergely Buday
46
Dichotomy results for constraint satisfaction problems
Michael Kompatscher
51
MSO model checking of graphs with unbounded rank-width
Eduard Eiben
53
Some Cardinal Invariants of the Generalized Baire Spaces
Diana Carolina Montoya Amaya
56
Gaps In ω ω
Marlene Koelbing
58
The lifting problem for measure and category
Anda Ramona Tănasie
60
Unified Correspondence as a Proof-Theoretic Tool
Apostolos Tzimoulis
62
Logical Metatheorems for Abstract Spaces axiomatized in Positive
Bounded Logic
66
Daniel Günzel
3
Proof Mining in Nonlinear Analysis
Daniel Körnlein
70
On the Why and How of implicit conflicts in Abstract Argumentation 72
Christof Spanring
Extracting consequence relations from abstract argumentation frames 76
Esther Anna Corsi
The Expressive Power of k-ary Inclusion-Exclusion Logic
Raine Rönnholm
79
4
Program
Thursday, May 14
8:30 Opening
8:45–9:30 Thomas Eiter
9:30–10:00 Break
10:00–10:30 Sylwia Polberg
10:30–11:00 Jesse Heyninck
11:00–11:30 Zhiguang Zhao
11:30–12:15 Thomas Eiter
12:15–13:30 Lunch break
13:30–14:30 Torsten Schaub
14:30–15:00 Matteo Pascucci
Querying Ontology Knowledge Bases
using Datalog
(p. 9)
Intertranslatability of Abstract Argumentation Frameworks
Combining Different Forms of Defeasible Reasoning in Abstract Argumentation: Integrating Mechanisms from
Adaptive Logics
An Abstract Algebraic Logic view of
propositional-attitude aggregation theory
Towards a Logic-Based Framework for
Analyzing Stream Reasoning
(p. 22)
Answer Set Solving in a nutshell
Basic model theory of modal languages
with propositional constants
15:00–15:30 Przemysław Andrzej Modal Logics for Qualitative Spatial
Wałęga
Reasoning
Algebra-based canonical formulas for
15:30–16:00 Julia Ilin
intermediate logics
16:00–16:30 Break
16:30–17:15 Uli Sattler
Complexity in Logic: off the beaten
track towards pastures new (1)
17:15–17:45 Wolfgang Fischl and Towards Reconciling SPARQL and CerShqiponja Ahmetaj tain Answers
17:45–18:15 Gergely Buday
Verifying the SECD machine in Nominal Isabelle
20:15
Unofficial event: Vienna Philharmonic
Summer Night Concert
(p. 26)
(p. 30)
(p. 10)
(p. 11)
(p. 33)
(p. 36)
(p. 40)
(p. 12)
(p. 43)
(p. 46)
5
Friday, May 15
9:00–9:45 Uli Sattler
9:45–10:15 Break
10:15–10:45 Michael Kompatscher
Complexity in Logic: off the beaten (p. 12)
track towards pastures new (2)
Dichotomy results for constraint satisfaction problems
10:45–11:15 Eduard Eiben
MSO model checking of graphs with
unbounded rank-width
11:15–11:30 VCLA International Student Awards Ceremony
11:30–12:00 VCLA International Student Awards: Sophie Spirkl
12:00–12:30 VCLA International Student Awards: Luke Schaeffer
12:30–14:00 Lunch break
14:00–14:45 Michael Moortgat
New directions in typelogical semantics
(1)
14:45–15:15 VCLA International Student Awards: Kuldeep S. Meel
15:15–15:45 VCLA International Student Awards: Pablo Munoz
15:45–16:15 Break
16:15–17:00 Michael Moortgat
New directions in typelogical semantics
(2)
17:00–17:30 Diana Carolina Montoya Some Cardinal Invariants of the GenerAmaya
alized Baire Spaces
17:30–18:00 Marlene Koelbing
Gaps in ω ω
18:00–18:30 Anda-Ramona Tanasie The lifting problem for measure and
category
Conference Dinner at Heuriger Werner20:00
Welser (http://www.werner-welser.at/)
(p. 51)
(p. 53)
(p. 17)
(p. 18)
(p. 13)
(p. 19)
(p. 20)
(p. 13)
(p. 56)
(p. 58)
(p. 60)
6
Saturday, May 16
9:45–10:45 Revantha Ramanayake Cut-elimination in generalisations of the
sequent calculus
10:45–11:15 Break
11:15–11:45 Apostolos Tzimoulis
Unified correspondence as a prooftheoretic tool
11:45–12:15 Daniel Günzel
Logical Metatheorems for Abstract
Spaces axiomatized in Positive Bounded
Logic
12:15–12:45 Daniel Koernlein
Proof Mining in Nonlinear Analysis
12:45–13:45 Lunch break
On the Why and How of implicit con13:45–14:15 Christof Spanring
flicts in Abstract Argumentation
14:15–14:45 Esther Anna Corsi
Extracting consequence relations from
abstract argumentation frames
14:45–15:15 Raine Rönnholm
The Expressive Power of k-ary
Inclusion-Exclusion Logic
15:15 Closing
(p. 14)
(p. 62)
(p. 66)
(p. 70)
(p. 72)
(p. 76)
(p. 79)
7
Invited Talks
Querying Ontology Knowledge Bases using
Datalog
Thomas Eiter
Knowledge Based Systems Group, TU Wien
Description logics play a dominant role in the formalization of ontologies for
practical applications of computer science, and they provide, for instance, the formal
underpinning of the Web Ontology Language (OWL) recommended by the World
Wide Web consortium (W3C). While description logics had been initially targeted
at reasoning about conceptual knowledge, in the last decade querying description
logic knowledge bases (often simply called ontologies) to elicit information about
individuals similar as from databases has received growing attention. Different
techniques for query answering have been proposed, among them also to query
rewriting to Datalog, which is a well-established query language for relational
databases that has several effective reasoning engines available. We consider particular
query types, instance and conjunctive queries, which for several description logics have
been transformed to Datalog, among them ones underlying the OWL 2 Profiles. These
transformations can also be beneficially exploited for extensions of ontologies, such
as in combinations with rules, and have led to research prototype implementations.
To exploit and extend the Datalog approach to query answering is an active area,
which provides a number of research opportunities.
9
Towards a Logic-Based Framework for
Analyzing Stream Reasoning
Thomas Eiter
Knowledge Based Systems Group, TU Wien
The rise of smart applications has drawn interest to logical reasoning over data
streams. Recently, different query languages and engines for stream processing
respectively stream reasoning were proposed in different communities. However,
due to a lack of theoretical foundations, the expressiveness and semantics of diverse
approaches was given only informally, and their semantic relationship. Towards
clear specifications and means for analytic study, a formal framework is desired
that allows to characterize their semantics in precise terms. Inspired by this, we
develop a logic-based such framework, which features window operators that provide
a flexible mechanism to represent views on streaming data and temporal operators
over windows, which also may be nested. In addition, we define a rule language on
top. We present the emerging formalism on examples and discuss some complexity
issues for it. Furthermore, we briefly consider its usage and relationship to stream
query respectively streams reasoning languages, in particular to the Continuous
Query Language (CQL) and ETALIS. We finally address recent results and ongoing
research around LARS, which is carried out in a project funded by the Austrian
Science Fund.
10
Answer Set Solving in a nutshell
Torsten Schaub
University Potsdam
The tutorial aims at acquainting the participant with ASP’s modeling and
solving methodology, enabling her/him to conduct independent problem solving
using ASP systems. To this end, the tutorial starts with a brief introduction to
the essential formal concepts of ASP, needed for understanding its semantics and
solving technology. In fact, ASP solving rests on two major components: A grounder
turning specifications in ASP’s modeling language into propositional logic programs
and a solver computing a requested number of answer sets of the program. We
illustrate both ASP’s grounding techniques and the basic ideas of the underlying
solving technology. Finally, we sketch the usage of ASP in conjunction with Python
for modeling complex reasoning scenarios. This involves an introduction to the
API of clingo 4, an ASP system with control capacities expressible in Python. We
illustrate this by developing the board game of Ricochet Robots.
All involved ASP systems are freely available from http://potassco.sourceforge.net.
11
Complexity in Logic: off the beaten track
towards pastures new
Uli Sattler
University of Manchester
Research in logic is hard: first, there are loads of existing formalisms, their
relationships, model properties, computational complexity classes, etc. that we have
to learn about and understand. Then, there is a wide range of skills to acquire,
including writing, note keeping, proof techniques, and analytical thinking. Of
course, we need to narrow down an interesting yet solvable research question and
make progress towards its solution. Many of these questions are basically “what
is the computational complexity of problem X in logic Y?”. In my tutorial, I will
first describe the above sources of complexity in more detail, and then present
some interesting cases where starting from but straying from this “beaten track” of
computational complexity has resulted in interesting insights. These cases include
new reasoning problems or tasks, and also alternative measures of complexity.
12
New directions in typelogical semantics
Michael Moortgat
Universiteit Utrecht
Typelogical grammars are substructural logics that provide a prooftheoretic
perspective on natural language syntax and semantics. In this tutorial I discuss
two recent developments in this area. The first (see [1] and much subsequent work)
is the integration of Montague-style compositional interpretation with a vectorbased account of lexical meanings. The second development ([2] and subsequent
work) systematically extends the vocabulary of Lambek’s original calculus: next to
composition (‘merge’) and its residuals, one adds decomposition and subtraction,
and a set of unary operators (related to the ‘!’ of linear logic) licensing limited forms
of reordering/restructuring that leave the form-meaning correspondence intact. I
will discuss the connection between these two developments and the opportunities
for transfer of results from one to the other.
References
[1] B. Coecke, M. Sadrzadeh, and S. Clark. Mathematical foundations for a compositional distributional model of meaning. CoRR, abs/1003.4394,2010.
[2] M. Moortgat. Symmetric categorial grammar. J. Philosophical Logic, 38(6):681–
710, 2009.
13
Cut-elimination in generalisations of the
sequent calculus
Revantha Ramanayake
TU Wien
Logic is concerned with the study and use of valid reasoning. The most wellknown logics are classical propositional and first-order logic. Nevertheless, various
other forms of reasoning are needed to model the different applications and situations
that arise in practice, giving rise to many new logics (‘non-classical logics’) that are
more expressive and permit finer distinctions than classical logic. The study of such
logics is interesting in its own right and also yields an understanding of the systems
modelled by these logics.
The basis of the proof-theoretical approach to studying a logic is the notion
of proof, and in particular, the study of the formal proof systems of the logic (a
statement belongs to the logic if and only if it has a proof in such a proof system).
Gentzen introduced the formal proof-system called the sequent calculus in order to
study the structure of proofs in classical and intuitionistic logic. His celebrated cutelimination theorem implies the sub-formula property which states that a statement
can be proved in the calculus using only sub-formulae of the formulae occurring in
the statement. The point is that the proofs from his calculus have a nice normal form
(such calculi are called analytic). Contrast, for example, with proofs in the Hilbert
calculus where formulae that occur in the proof might not occur in the statement
that is proved. Gentzen made use of his result to give a formal proof of consistency
of Peano arithmetic using a suitable induction principle.
Note that if we are not interested in having a calculus with the subformula
property, then it is easy to obtain a sequent calculus for most logics of interest.
However the key to a proof-theoretical study of the logic is an analytic calculus.
For many logics of interest it is not clear at all how to obtain an analytic sequent
calculus. This has led to the introduction of various generalisations and extensions of
the sequent calculus. These generalisations yield analytic calculi for some logics and
not for others and it is still not clear why cut-elimination seems to fail for certain
logics in certain formal systems (negative results stating that a given formal system
cannot yield an analytic calculus for a certain logic are rare!).
In this tutorial I will discuss two generalisations of the sequent calculus: the
hypersequent calculus and the Display Calculus. I will pay special attention to
the display calculus which has been used to provide analytic formal systems for
many different non-classical logics. An attractive feature of the display calculus is
the natural general cut-elimination theorem which applies whenever the rules of
14
the calculus satisfy certain easy to check conditions. I will ‘take apart’ the display
calculus in order to motivate why the rules of calculus take the shape that they do,
and why this leads to a general cut-elimination theorem. I will then explain how a
display calculus for a logic can be extended by suitable new rules to obtain a display
calculi for axiomatic extensions of that logic. The preservation of analyticity under
the addition of new rules is known as modularity and this property usually fails in
the sequent calculus but holds for the rules of the display calculus we introduce here.
An intention of this tutorial is that the insights on proof systems and cut-elimination
explained via the display calculus will be applicable to various other proof systems
as well.
15
Winners of the
VCLA
International
Student Awards
Boolean Circuit Optimization
Sophie Spirkl
Princeton University
We consider the problem of constructing adders with prescribed input arrival
times. Most previous results implement parallel prefix graphs (e.g. Kogge-Stone)
and are designed for uniform input arrival times. We generalize the concept of prefix
graphs, which allows us to reduce single-output adder optimization problems to a
tree structure, and we allow arbitrary input arrival times. For both single-output
and full adders, we present efficient algorithms which construct adders that improve,
even for uniform arrival times, upon previous results in the core objectives delay,
size and fan-out.
17
Deciding Properties of Automatic Sequences
Luke Schaeffer
MIT
Automatic sequences are self-similar sequences, defined in terms of finite automata,
which arise naturally as the solutions to pattern avoidance problems. We show that
several questions about automatic sequences can be expressed as logical predicates in
a decidable first-order theory, and then answered purely mechanically by a decision
algorithm. With the aid of this algorithm, we recover a surprising number of known
results, and prove interesting new theorems as well. We also improve the theoretical
power of this approach by extending the logical theory to broader classes of sequences,
adding new operations, and showing how to better interpret the results.
18
SAT Sampling: From Theory to Practice
Kuldeep S. Meel
Rice University
Counting the number of satisfying truth assignments of a given Boolean formula
or sampling such assignments uniformly at random are fundamental computational
problems in computer science with numerous applications. In computer-aided design,
these problems come up in constrained-random verification, where test input vectors
are described by means of constraints. While the theory of these problems has
been thoroughly investigated in the 1980s, approximation algorithms developed by
theoreticians do not scale up to industrial-sized instances. Algorithms used by the
industry offer better scalability, but give up certain correctness guarantees to achieve
scalability. We describe a novel approach, based on universal hashing and SMT,
that scales to formulas with hundreds of thousands of variables without giving up
correctness guarantees.
19
New Complexity Bounds for Evaluating
CRPQs with Path Comparisons
Pablo Munoz
University of Chile
Graph databases make use of logics that combine traditional first order features
with navigation on paths, in the same way logics for model checking do. However,
modern applications of graph databases impose a new requirement on the expressiveness of the logics: they need comparing labels of paths based on word relations. This
has led to the study of logics that extend basic graph languages with features for
comparing labels of paths based on regular relations, or the strictly more powerful
rational relations. The evaluation problem for the former logic is decidable (and even
tractable in data complexity), but already extending this logic with such a common
rational relation as subword or suffix turns evaluation undecidable. We thus study
less expressive logics that still allow comparing paths based on practically motivated
rational relations. Here we concentrate on the most basic such languages, which
extend graph pattern logics with path comparisons based only on suffix, subword
or subsequence. The results provide a complete landscape of the complexity of
evaluation for each one of these logics, which are all decidable in elementary time.
The extension with suffix is even tractable in data complexity, while the other two
are not. To obtain our results we establish a link between the evaluation problem
for graph logics and two important problems in word combinatorics: word equations
with regular constraints and square unshuffling.
20
Contributed
Talks
Intertranslatability of Abstract
Argumentation Frameworks
Sylwia Polberg
Over the last years, argumentation has become an influential field in Artificial
Intelligence [1]. One of its subfields is abstract argumentation, at the heart of which
lies the abstract argumentation framework (AF) developed by Phan Minh Dung [2].
Although well acknowledged, AFs have their shortcomings, which inspired a search
for more general models [3]. Throughout the years, many AF extensions were created,
ranging from the ones employing various values and preferences [4, 5, 6] to those that
focus on researching new types of relations between arguments [7, 8, 9, 10, 11, 12].
This amount of frameworks should not come as a surprise. Argumentation is a
wide area with numerous applications, in which one has to face different classes
of problems. Frameworks of a given type can be seen as tools to model particular
issues and concepts, which on one hand gives us more insight into how to approach
the problems, but on the other affects the framework’s design. When facing such
amount of available structures, it is only natural to ask whether one can translate one
framework into another, particularly in a way that preserves the desired semantics,
and what the consequences of such a process are.
Framework intertranslatability is an interesting topic from both the practical and
theoretical point of view. When it comes to applications, establishing a transformation
can be useful for the design of argumentation–based software. While modeling
problems in more advanced frameworks has its benefits, many structures do not have
dedicated solvers. Consequently, translating a framework into one that is supported
by a software implementation is of practical value. Moreover, based on available
translations and their complexity, we can choose structures optimal for further solver
development. Finally, if a translation of a given framework produces a particular
subclass of the target structure, we can use this knowledge to further improve the
targets solver.
Translations are often used to compare the expressive power of given formalisms.
While in the application–oriented approach we first and foremost look for an approach
that would produce us the desired extensions, here we search for transformations
that exhibit particular properties such as faithfulness, polynomiality and modularity
[13]. Additionally, in case we cannot find any, we try to establish impossibility results
proving that they in fact cannot exist. However, our research interests go beyond
the expressive power. First of all, we search for “generic” translations, i.e. ones
that are as independent from the argumentation semantics as possible. Although
dedicated transformations can use various properties of given semantics to e.g. boost
22
efficiency, generic ones are more useful in exposing and comparing the design choices
and capabilities of the frameworks we are interested in. Furthermore, we want to
establish the connection between various relations between arguments, in particular
between different forms of support and attack. We would like to know if it is possible
to transform one into another and what is lost or gained in the process, thus providing
an answer to the question what is the value of researching additional relation types
in abstract argumentation frameworks.
In our study we introduce a wide range of translations between the aforementioned
frameworks that research new relations between arguments [2, 7, 8, 9, 10, 11, 12].
Some methods have already been researched, especially transformations from and to
the Dung framework. We extend these results by providing a way to transform the
abstract dialectical framework into the Dung framework which so far has been done for
only several semantics [14]. Furthermore, the existing methods for frameworks with
support [15, 16, 10] can be classified as coalition approaches, i.e. the arguments in the
target structure represent sets of arguments of the source structure that are connected
by support. We propose two alternative transformations – the defender and attack
propagation – inspired by the research in [17, 8], which aim to simulate the behavior
of support with combinations of attack and defense. Finally, despite the available
research, there are less results concerning moving between AF generalizations [8].
Consequently, we introduce new approaches and follow up on our previous work [11].
We also provide an in–depth analysis of our translations in terms of functional,
complexity, syntactical and semantical properties. The first group studies the
translation as a function and answers questions such as: can all source frameworks
be translated, can any target framework be obtained, and whether there is a one–
to–one relation between them. The complexity properties focus on how difficult
the translation is in terms of computational time, modularity, required semantical
knowledge of a given framework and the size of the translated structure relative to
the source one. Syntactical properties tell us about the difference in the domains of
arguments and introduction or removal of arguments and relations between the source
and target frameworks. Finally, the last group compares the semantics behavior of
both structures, i.e. if all the desired extensions and only them can be obtained, are
auxiliary arguments required and whether the translation is specialized for particular
semantics or is generic and can be applied in more cases. This analysis allows us to
separate available argumentation frameworks into groups, inside of which we can
“move” between structures at a low cost. This separation can be seen as a rough
classification of expressive power of available argumentation frameworks.
To summarize, we introduce a wide range of translations between argumentation
frameworks and analyze their properties in order to compare and analyze the
frameworks. However, the purpose of our research is not to advocate the use of one
relation or framework above the other. By the means of translations we want to
show the consequences of changing between different structures with different types
and amounts of elements. The consequences which, depending on the problems we
23
want to model, can be considered very heavy or completely negligible. Therefore, our
classification and comparison can be seen as an aid in choosing tools adequate for our
purposes, while translations can be used to limit some of their possible drawbacks.
References
[1] Bench-Capon, T.J.M., Dunne, P.E.: Argumentation in artificial intelligence.
Artif. Intell. 171(10-15) (2007) 619–641
[2] Dung, P.M.: On the acceptability of arguments and its fundamental role in
nonmonotonic reasoning, logic programming and n-person games. Artif. Intell.
77 (1995) 321–357
[3] Brewka, G., Polberg, S., Woltran, S.: Generalizations of Dung frameworks and
their role in formal argumentation. Intelligent Systems, IEEE 29(1) (Jan 2014)
30–38
[4] Amgoud, L., Vesic, S.: A new approach for preference-based argumentation
frameworks. Ann. Math. Artif. Intell 63 (2011) 149–183
[5] Bench-Capon, T.J.M.: Persuasion in practical argument using value-based
argumentation frameworks. J. Log. Comput. 13(3) (2003) 429–448
[6] Modgil, S.: Reasoning about preferences in argumentation frameworks. Artif.
Intell. 173(9-10) (2009) 901–934
[7] Baroni, P., Cerutti, F., Giacomin, M., Guida, G.: AFRA: Argumentation
framework with recursive attacks. Int. J. Approx. Reasoning 52(1) (2011) 19–37
[8] Cayrol, C., Lagasquie-Schiex, M.C.: Bipolarity in argumentation graphs: Towards a better understanding. Int. J. Approx. Reasoning 54(7) (2013) 876–899
[9] Nielsen, S., Parsons, S.: A generalization of Dung’s abstract framework for
argumentation: Arguing with sets of attacking arguments. In: Proc. ArgMAS.
Volume 4766 of LNCS. Springer (2007) 54–73
[10] Nouioua, F.: AFs with necessities: Further semantics and labelling characterization. In Liu, W., Subrahmanian, V., Wijsen, J., eds.: Proc. SUM ’13. Volume
8078 of LNCS. Springer Berlin Heidelberg (2013) 120–133
[11] Polberg, S., Oren, N.: Revisiting support in abstract argumentation systems.
In: Proceedings of COMMA 2014. (2014)
[12] Brewka, G., Woltran, S.: Abstract dialectical frameworks. In: Proc. KR ’10,
AAAI Press (2010) 102–111
24
[13] Janhunen, T.: On the intertranslatability of non–monotonic logics. Annals of
Mathematics and Artificial Intelligence 27(1-4) (1999) 79–128
[14] Brewka, G., Dunne, P.E., Woltran, S.: Relating the Semantics of Abstract
Dialectical Frameworks and Standard AFs. In: Proceedings of the 22nd International Joint Conference on Artificial Intelligence (IJCAI 2011), AAAI Press
(2011) 780–785
[15] Cayrol, C., Lagasquie-Schiex, M.C.: Coalitions of arguments: A tool for handling
bipolar argumentation frameworks. Int. J. Intell. Syst. 25(1) (2010) 83–109
[16] Oren, N., Reed, C., Luck, M.: Moving between argumentation frameworks.
In: Proceedings of the 2010 conference on Computational Models of Argument:
Proceedings of COMMA 2010, Amsterdam, The Netherlands, The Netherlands,
IOS Press (2010) 379–390
[17] Cayrol, C., Lagasquie-Schiex, M.C.: Bipolar abstract argumentation systems.
In Simari, G., Rahwan, I., eds.: Argumentation in Artificial Intelligence. (2009)
65–84
25
Combining Different Forms of Defeasible
Reasoning in Abstract Argumentation:
Integrating Mechanisms from Adaptive Logics
Jesse Heyninck
In this talk I will present a framework that allows for the combination of different
forms of defeasible reasoning (in short, DR).
DR (e.g. inconsistency handling, inductive generalizations, abduction, normative
reasoning, etc.) is an essential form of reasoning whenever encountering complex,
uncertain or incomplete information. When we reason defeasibly, we draw conclusions
only tentatively, i.e. they may be retracted in the light of new information. This
means that the support of a defeasible argument is not deductive: it is possible that
the premises of a valid defeasible argument are true while its conclusion is false.
The conclusion is thus drawn conditionally, and when the assumptions under which
the conclusion is drawn turn out to be dubious, we retract this inference. We can
thus say that DR always consists of an ampliative aspect (i.e. the fact that there
is more derivable then what is deductively guaranteed) and a corrective aspect (i.e.
the fact that conclusions can be retracted if they turn out to be based on dubious
assumptions). This also makes clear that DR has an essentially dialectical character:
defeasible inferences are assumed valid until and unless encountering an argument
against their validity.
A formal model that allows for the systematic study of the dialectical character
of DR is abstract argumentation, developed in the seminal paper by Dung [5]. In
this framework arguments are arranged in directed graphs hA, →i in which A is a
set of arguments and → ⊆ A × A represents argumentative attacks (e.g. Pollock’s
rebuttals and undercuts). Given such a graph, argumentation semantics specify
criteria for selecting sets of arguments that represent stances of rational discussants.
In recent years, however, abstract argumentation has received some criticism for
lacking an account of the logical structure of argumentation (see for example [3]
and [10]). Not only is the logical structure of argumentation an essential part of
the practice of reasoning, there are also some more specific problems caused by the
high level of abstraction on which abstract argumentation operates. For example,
abstract argumentation semantics in general fail to fulfil certain intuitive rationality
constraints such as the consistency or the logical closure of selected conclusions.
Several proposals to add more structure to AFs have been made in the literature
(see for example: [1], [9], [8], [7], [6]). One problem of these approaches is, however,
that within the logical models they depend on, the mechanisms of specific instances
26
of DR frequently used in science and the interplay of different forms of DR has not
been studied in detail.
A good starting point to improve on these approaches is to integrate mechanisms
from a formal framework within which many different DR forms have been modelled.
Such a framework is given by adaptive logics. The standard format for adaptive
logics [2, 11] offers a generic framework for DR. Its dynamic proof theory extends a
monotonic, reflexive and transitive core logic (L) with a set of retractable inferences
which are associated with defeasible assumptions. More specifically, these assumptions
are sets of formulas ∆ of a predefined ‘abnormal’ form that are assumed to be false
in the given inference. When an assumption turns out to be dubious in view of a
premise set Γ, e.g. when some A ∈ ∆ is derived from Γ in L as part of a minimal
disjunction of abnormalities, the inference associated with it gets retracted. Various
adaptive strategies offer mechanisms for this retraction of inferences, some following
a more cautions rationale then others. A plethora of forms of defeasible reasoning
has been explicated in the adaptive logic framework.
Examples are inconsistency-adaptive logics based on the paraconsistent core logic
CLuN where abnormalities are contradictions A ∧ ¬A. One retractable inference
rule is disjunctive syllogism (DS): ¬A, A ∨ B implies B on the assumption that
A ∧ ¬A is false. Take the premise set Γ = {¬p, p, ¬r, p ∨ s, r ∨ q}. While applying
DS to ¬r and r ∨ q, assuming that r ∧ ¬r is false, is not retracted, applying DS to
¬p and p ∨ s will be retracted since p ∧ ¬p is derivable.
In many real life examples of DR, different forms of DR are used in combination
with one another. Only recently have the first steps been taken in the direction of
a framework that allows for the combination of different sorts of adaptive logics
(see [12], [13], [11, Ch.5] and [4]). In contrast to the standard format of adaptive
logics, in these approaches there is no transparent account of the dialectical character
of defeasible inferences. Also, the field of adaptive logics still stands in relative
isolation from approaches to defeasible reasoning in the argumentative tradition,
such as ASPIC [9], Defeasible Logic Programming [7] and argumentation based
on classical logic [6]. This isolation inhibits both the comparative studies and the
cross-fertilization between these two approaches.
In this talk I present a framework for the representation of combinations of DR
forms in formal argumentation. In this framework, arguments are generated by
(possibly multiple) stable core logics and the retractable inference rules extending
them. Argumentative attacks are then defined to represent the mechanism that
retracts inferences based on dubious assumptions. This framework allows us to
integrate the well-known mechanisms of ALs for the modelling of DR forms . It
retains a clear account of the dialectical nature of DR without excluding expressiveness. I give representation results for the standard, lexicographic and colexicographic
format of AL. Furthermore, I give a representation result for ASPIC, one of the
best known frameworks for structured argumentation. The representational results
of the lexicographic and colexicographic format of adaptive logic are used to show
27
the potential to model combinations of DR forms. I end the talk by pointing to
promising future research based on this framework that could be beneficial both to
the adaptive logic and the formal argumentation communities.
References
[1] L. Amgoud and P. Besnard. A formal analysis of logic-based argumentation
systems. In SUM, pages 42–55, 2010.
[2] D. Batens. A universal logic approach to adaptive logics. Logica Universalis,
1:221–242, 2007.
[3] M. Caminada and Y. Wu. On the limitations of abstract argumentation. In
Proceedings of the 23rd Benelux Conference on Artificial Intelligence (BNAIC),
Gent, Belgium, 2011.
[4] S. Christian and F. Van De Putte. Proof theories for superpositions of adaptive
logics. Logique et Analyse, Forthcoming.
[5] P. M. Dung. On the acceptability of arguments and its fundamental role in
nonmonotonic reasoning, logic programming and n-person games. Artifical
Intelligence, 77:321–358, 1995.
[6] N. Gorogiannis and A. Hunter. Instantiating abstract argumentation with
classical logic arguments: Postulates and properties. Artifical Intelligence,
175(9-10):1479–1497, 2011.
[7] G. Governatori and M. J. Maher. An argumentation-theoretic characterization of
defeasible logic. In W. Horn, editor, Proceedings of the 14th European Conference
on Artificial Intelligence, pages 469–474, Amsterdam, 2000. IOS Press.
[8] H. Prakken. An argumentation framework in default logic. Annals of Mathematics and Artificial Intelligence, 9(1):93–132, 1993.
[9] H. Prakken. An abstract framework for argumentation with structured arguments. Argument and Computation, 1(2):93–124, 2010.
[10] H. Prakken. Some reflections on two current trends in formal argumentation. In
A. A. R. Craven, N. K. Çiçekli, and B. S. K. Stathis, editors, Logic Programs,
Norms and Action, pages 249–272. Springer, 2012.
[11] C. Straßer. Adaptive Logic and Defeasible Reasoning. Applications in Argumentation, Normative Reasoning and Default Reasoning. Springer, 2014.
[12] F. Van De Putte and C. Straßer. Extending the standard format of adaptive
logics to the prioritized case. Logique et Analyse, 220:601–641, 2012.
28
[13] F. Van De Putte and C. Straßer. Three formats of prioritized adaptive logics: a
comparative study. Logic Journal of IGPL, 21(2):127–159, 2013.
29
An Abstract Algebraic Logic view of
propositional-attitude aggregation theory
María Esteban and Zhiguang Zhao
The theory of social choice is the formal study of mechanisms for collective
decision making, and investigates issues of philosophical, economic, and political
significance, stemming from the classical Arrovian problem of how the preferences of
the members of a group can be “democratically” aggregated into one outcome.
In the last decades, many results appeared generalizing the original Arrovian
problem (e.g. abstract and algebraic aggregation theory by Wilson [16] and Rubinstein
and Fishburn [15]), which gave rise to a research area called judgment aggregation (JA)
[13], which studies how a group of individuals aggregate their individual judgments
on logically interconnected propositions (the agenda) into collective judgments on
them. While the original work of Arrow [1] focuses on preference aggregation, this
can be recognized as a special instance of a consistent judgments, expressed by
each member of a group of individuals over a given set of logically interconnected
propositions: each proposition in the agenda is either accepted or rejected by each
group member, so as to satisfy certain requirements of logical consistency. Within the
JA framework, the Arrovian-type impossibility results (providing sufficient axioms
for aggregator functions to turn into degenerate rules, such as dictatorship) are
obtained as consequences of characterization theorems [14], which provide necessary
and sufficient conditions for agendas to have aggregator functions on them satisfying
given axiomatic conditions.
In the same logical vein, in [12], attitude aggregation theory was introduced;
this direction has been further pursued by Herzberg [8], where a characterization
theorem has been given for certain many-valued propositional-attitude aggregators
as MV-algebra homomorphisms, and a well-known result from judgment aggregation
has been derived from it as corollary.
Methodologically, the ultrafilter argument (see Kirman and Sondermann [10],
Lauwers and van Liedekerke [11]) is the tool, derived from logic, underlying the
generalizations and unifications mentioned above. It can be sketched as follows: to
prove impossibility theorems for finite electorates, one shows that the axiomatic
conditions on the aggregation function force the set of all decisive coalitions to be an
(ultra)filter on the powerset of the electorate. If the electorate is finite, this implies
that all the decisive coalitions must contain one and the same (singleton) coalition:
the oligarchs (the dictator). First employed in the proof of Arrow’s theorem, this
argument was applied to obtain elegant and concise proofs of impossibility theorems
in judgment aggregation [3]. More recently, it gave rise to characterization theorems,
30
e.g. establishing a bijective correspondence between Arrovian aggregation rules and
ultrafilters on the set of individuals [9]. Moreover, by means of the well-known
correspondence between ultrafilters and Boolean homomorphism, such argument
has been generalized by Herzberg to get a bijective correspondence between certain
judgment aggregation functions and ultraproducts of profiles [7].
The present paper is motivated by the observation that Abstract Algebraic Logic
(AAL) is the natural theoretical setting for Herzberg’s results.
Abstract Algebraic Logic (AAL) is a forty-years old research field in mathematical
logic (see [5], [6]). It was conceived as the framework for an algebraic approach to the
investigation of logics: its main goal was establishing a notion of canonical algebraic
semantics uniformly holding for classes of logics, and using it to systematically
investigate properties of logics in connection with properties of their algebraic
counterparts.
In the proposed talk, we report the results in the working paper [4], in which
an Abstract Algebraic Logic perspective is given on propositional attitude aggregation theory. Specifically, we generalize and refine Herzberg’s result in [8] from
the MV-algebra setting to any class of algebras canonically associated with some
selfextensional logic. This notion encompasses a vast class of logics, of which classical,
intuitionistic, modal, many-valued and relevance logic are special cases.
In particular, we improve Herzberg’s characterization result which consisted of
two slightly asymmetric parts. The present characterization result is symmetric, and
holds for the class of algebras canonically associated with any selfextensional logic
S. Aggregation of propositional attitudes modeled in classical, intuitionistic, modal,
many-valued and relevance logic can be uniformly captured as special cases of the
present result. This result paves the way to the systematic study of a wide array of
“realistic agendas” of formulas the propositional connectives of which are interpreted
in ways which depart from the classical interpretation. Conceptually, adopting
nonclassical logics to formalize natural language statements makes it possible to
more adequately capture the original meaning in different situations. Technically,
the nonclassical interpretation of the logical connectives is useful in the light of
the fact that, as observed by Dietrich [2], it can provide a strategy to escape the
impossibility results (for instance, in the case of [2], Dietrich escaped the impossibility
results by interpreting conditional propositions of the form “if a then b” by means of
subjunctive implication rather than classical implication).
References
[1] K. J. Arrow. Social choice and individual values, volume 12. John Wiley, New
York, 2nd edition edition, 1963.
[2] F. Dietrich. The possibility of judgment aggregation on agendas with subjunctive
implications. Journal of Economic Theory, 145(2):603–638, March 2010.
31
[3] F. Dietrich and P. Mongin. The premiss-based approach to judgment aggregation.
Journal of Economic Theory, 145(2):562–582, 2010.
[4] M. Esteban, A. Palmigiano, and Z. Zhao. An abstract algebraic logic view of
propositional-attitude aggregation theory. in preparation, 2015.
[5] J. M. Font and R. Jansana. A General Algebraic Semantics for Sentential Logics,
volume 7 of Lectures Notes in Logic. The Association for Symbolic Logic, Ithaca,
N.Y., second edition, 2009.
[6] J.M. Font and R. Jansana. A general algebraic semantics for sentential logics.
Lecture notes in logic. Springer-Verlag, 1996.
[7] F. Herzberg. Judgment aggregation functions and ultraproducts. Institute of
Mathematical Economics, University of Bielefeld, 2008.
[8] F. Herzberg. Universal algebra for general aggregation theory: Many-valued
propositional-attitude aggregators as MV-homomorphisms. Journal of Logic
and Computation, 2013.
[9] F. Herzberg and D. Eckert. Impossibility results for infinite-electorate abstract
aggregation rules. Journal of Philosophical Logic, 41:273—286, 2012.
[10] Alan P Kirman and Dieter Sondermann. Arrow’s theorem, many agents, and
invisible dictators. Journal of Economic Theory, 5(2):267–277, 1972.
[11] Luc Lauwers and Luc Van Liedekerke. Ultraproducts and aggregation. Journal
of Mathematical Economics, 24(3):217–237, 1995.
[12] C. List and F. Dietrich. The aggregation of propositional attitudes: towards a
general theory. In Oxford Studies in Epistemology, volume 3, pages 215–234.
Oxford University Press, 2010.
[13] C. List and B. Polak. Introduction to judgment aggregation. Journal of
Economic Theory, 145(2):441 – 466, 2010.
[14] K. Nehring and C. Puppe. Strategy-proof social choice on single-peaked domains:
Possibility, impossibility and the space between. University of California at
Davis, 2002.
[15] Ariel Rubinstein and Peter C. Fishburn. Algebraic aggregation theory. Journal
of Economic Theory, 38(1):63–77, 1986.
[16] Robert Wilson. On the theory of aggregation. Journal of Economic Theory,
10(1):89–99, 1975.
32
Basic model theory of modal languages with
propositional constants
Matteo Pascucci
1
Introduction
From a semantic perspective a propositional modal language LM OD can be viewed
either as a fragment of a first-order language L1 , when interpreted in models, or as
a fragment of a monadic second-order language L2 , when interpreted in frames [5].
Here I will examine a modal language enriched with a set of propositional constants
CON = {c1 , c2 , ...}, which are symbols for special propositions, each introduced with
a set of characteristic axioms. These axioms define the meaning of propositional
constants, i.e. are used to restrict the set of their possible interpretations in a frame
and are not closed under uniform substitution. Furthermore, axioms for constants
can be expressed either in the modal language itself or in a metalanguage, for instance
L1 or L2 ; I will be concerned only with a language where all constants are “internally
axiomatized” (i.e. within the modal language), which can be called LcM OD . Some
examples of modal languages enriched with propositional constants of various kind
may be found in [1], [2], [3] and [4].
2
Frames and models
LcM OD -formulas are interpreted in general frames with specific restrictions. A general
frame with specific restrictions is G = hW, R, Π, {Π(ci )|ci ∈ CON }i, where Π ⊆ 2W
is the set of admissible interpretations for propositional variables (the restriction
for propositional variables) and Π(ci ) ⊆ Π is the set of admissible interpretations
for the i-th constant in the language (the restriction for the i-th constant). Notice
that Π(ci ) is not required to be closed under boolean and modal operations. An
admissible model on these frames is M = hG, V i, where V is a valuation function
such that V (pj ) ∈ Π for any variable pj and V (ci ) ∈ Π(ci ) for any constant ci .
Satisfiability and validity of LcM OD -formulas in admissible models and frames are
defined accordingly.
3
The proper range of a formula
The notion of proper range will be used to distinguish a certain class of models for a
formula. Let Ck be the class of all general frames (with specific restrictions) whose
33
domain has cardinality k. We say that G = hW, R, Π, {Π(ci )|ci ∈ CON }i ∈ Ck has
the weakest restrictions of level k for an LcM OD -formula ψ when:1
• G ψ;
• |Π| ≥ |Π0 |, for any restriction Π0 s.t. there is G0 [Π0 ] ∈ Ck and G0 [Π0 ] ψ;
• for any Π0 ⊃ Π, there is no G0 [Π0 ] ∈ Ck s.t. G0 [Π0 ] ψ;
• for any constant ci ∈ ψ, |Π(ci )| ≥ |Π0 (ci )|, for any specific restriction Π0 (ci ) s.t.
there is G0 [Π0 (ci )] ∈ Ck and G0 [Π0 (ci )] ψ;
• for any Π0 (ci ) ⊃ Π(ci ), there is no G0 [Π0 (ci )] ∈ Ck s.t. G0 [Π0 (ci )] ψ.
The proper range of ψ is defined as the smallest class including:
• for any cardinal k, all general frames with the weakest restrictions of level k
for ψ;
• the disjoint union of any family of frames already in the proper range of ψ.
This definition has some relevant analogous for languages without constants: indeed,
it can be proved that the proper range of an LM OD -formula φ corresponds to the
class of all models based on Kripke frames where φ is valid, or, equivalently, to the
class of all models built on general frames where φ is valid and Π = 2W .
4
Invariance and preservation results
It is well-known that the satisfiability of an LM OD -formula in a model is invariant
under generated submodels, disjoint unions, p-morphic images and ultrafilter extensions, whereas the validity of an LM OD -formula in a frame is preserved under
generated subframes, disjoint unions and p-morphic images and anti-preserved under
ultrafilter extensions [5]. It can be proved that for LcM OD -formulas the situation is
analogous. However, the expressivity of LM OD and LcM OD is different; for instance,
there is an LcM OD -formula φ containing two inverse operators of necessity and a
constant c which has for its proper range a class of irreflexive general frames. In
particular, we will see that:
• φ is valid in the general frame Z = hZ, <, Π, Π(c)i, where Z is the set of integers,
< is the relation “smaller than”, Π = 2Z and Π(c) = {{n}|n ∈ Z};
• there is no p-morphic image of Z containing a reflexive point.
1
Expressions like G[Π] are shorthands to say that Π is the restriction in G.
34
LcM OD differs in expressivity also from a modal language with nominals; for instance,
in [2] it is proved that the validity in a frame of a modal formula containing nominals
is not preserved under disjoint unions and p-morphic images. The reason is that
nominals can be seen as a particular kind of propositional constants which are not
“internally axiomatized”, since some of their intended properties need to be expressed
in L1 or L2 .
References
[1] A. R. Anderson. A reduction of deontic logic to alethic modal logic. Mind, vol.
67 (1958), pp. 100-103.
[2] P. Blackburn. Nominal tense logic. Notre Dame Journal of Formal Logic, vol. 34
(1992), pp. 56-83.
[3] C. Pizzi. A logic of contingency with a propositional constant. In Logic and
Philosophy in Italy, E. Ballo and M. Franchella (eds.), Polimetrica, Milano, 2006,
pp. 141-151.
[4] A. N. Prior. Papers on Time and Tense. Oxford University Press, 1968.
[5] J. van Benthem. Modal Logic and Classical Logic. Bibliopolis, Napoli, 1983.
35
Modal Logics for Qualitative Spatial
Reasoning
Przemysław Andrzej Wałęga
Space is an inexhaustible source of inspiration that for ages has fascinated
mathematicians, logicians and philosophers, among others. In this note, we are
mainly focused on formal methods for space representation and reasoning. Such
formalisms have been recently thoroughly studied by researchers from the fields of
Artificial Intelligence (AI) and Knowledge Representation (KR), as we can find out
from the preface of [1]:
’Space, with its manifold layers of structure, has been an inexhaustible
source of intellectual fascination since Antiquity. [. . . ] In this long intellectual history, however, one relatively recent, yet crucial, event stands
out: the rise of the logical stance in geometry. Fundamental to this
development is the analysis of geometrical structures in relation to the
formal languages used to describe them, and the recognition of the special
mathematical challenges – and opportunities – which such an analysis
presents.’
What is interesting from the cognitive and philosophical points of view is that
humans possess the abilities to reason about space that can hardly be obtained
by means of any AI system. Until now, no AI method has been able to perform
as precise and universal spatial reasoning as humans. This situation motivates
us to work on spatial reasoning methods that would imitate human-like reasoning
aspects. In particular, human-like methods seem to be based on qualitative approach
being in opposition to quantitative (numerical) approach. We are convinced that
using qualitative methods will be an important step towards better understanding
of human spatial reasoning methods. Additionally, this approach is usually faster,
has lower computational complexity and is easier to understand for humans than
the quantitative approach. As a result, the qualitative approach has a number of
practical spatial applications, e.g., in geographical information systems [2] or in
robotics [3].
There are three main formal approaches to qualitative spatial reasoning, namely
relational algebras, first-order theories and modal logics. Relational algebras (e.g.,
see [9]) deal with objects and binary, jointly exhaustive and mutually disjoint
relations between them. These purely existential theories are formulated as constraintsatisfaction systems, i.e., they check whether a given set of objects and relations
36
between them is consistent. Within the second approach (e.g., see [8]) various firstorder spatial theories of such spatial concepts as topology or directional information
are considered. Finally, there are spatial modal logics (e.g., see [4]) that introduce
modalities to represent spatial relations between objects. The last approach makes
it possible to obtain very expressive systems, but in many cases they have high
computational complexity (often they are even undecidable).
The first part of the presentation introduces the motivation for our work on
spatial formalisms. Afterwards, we confine ourselves to the description of qualitative
spatial modal logics and in particular, to two-dimensional logics for directional
information representation. We present Compass Logic [10], Spatial Propositional
Neighborhood Logic (SpPNL) [7] and Cone Logic [6]. The abovementioned logics
use the two-dimensional Cartesian coordinate system in which a point is identified
with a pair of numbers, namely its x and y coordinates. In this environment, various
modal operators that represent directional relations between points (or regions in
the case of SpPNL) are introduced.
• Compass Logic [10] involves two irreflexive linear orders for two Cartesian
coordinates. The first of them is to be interpreted as “lying horizontally”
and the second as “lying vertically” on a Cartesian grid. Four modalities are
introduced that enable us to move along one of the Cartesian axes keeping the
other coordinate constant. In other words, the modalities provide an access to
points lying on the same horizontal (or vertical) line as the given point. The
logic is proven to be undecidable [5].
• Spatial Propositional Neighborhood Logic [7] is another logic with projection
modalities that uses two linear orders for Cartesian coordinates. However,
instead of points (as in Compass Logic), SpPNL considers two-dimensional
regions. Four modalities are introduced which enable us to move along x and
y axes of the Cartesian coordinate system. SpPNL is also undecidable.
• Cone Logic [6] is not a projection-based spatial logic like the two previously
mentioned systems. Instead, it is based on the cone-shaped cardinal directions.
Given a particular point, it divides a plane into four regions: lower-left, lowerright, upper-left and upper-right quadrants. Afterwards, eight modalities are
introduced that make it possible to access points located in these quadrants.
Four of the modalities treat quadrants as open regions, whereas other four
treat quadrants as semi-closed regions. The logic is decidable and proven to be
PSpace-complete.
The abovementioned, well-known directional modal logics capture space from
the aerial point of view. This way of space description is appropriate, e.g., for
maps’ representation but it does not reflect the subject’s perspective and the way
she perceives the surrounding space. On the other hand, the subject-oriented
representation may be used as an interface between machine and humans in a sense
37
that qualitative spatial information expressed by a human would be interpreted by a
machine and the other way round. As a result, such interfaces may be applied in
systems that describe surrounding space and navigate a blind person. Systems with
subject-oriented representation may also be used in mobile robotics for autonomous
navigation. In order to capture the subject-oriented representation we propose a
novel modal language called HLQL. We present a new type of directional operators
that represent information about object’s position with respect to another object
from a subject’s point of view. These operators involve 3 elements (in standard
approaches only 2 elements are considered) and enable us to represent such relations
as: “from my current point of view, the post office is to the left of the church”.
Technically,
• HLQL is a basic hybrid multi-modal language (involving nominals and satisfaction operators) augmented with appropriately tailored accessibility relations
and the constant symbol s (for the subject). The semantics for the logic is
Kripke-structure based: a frame is a plane, either finite or infinite, with polar
coordinates, divided into cells of arbitrary length and angle-width. The central
locus is occupied by the subject. We introduce modal operators that make it
possible to express directional position of an object with respect to the subject,
an object’s location with respect to another object, seen from the subject’s
perspective, and also qualitative distance relations between objects. We have
proven that the axiomatization of HLQL over finite domains is sound and
complete with respect to the indicated class of structures. For finite domains
the logic is decidable.
In our future work we would like to devise a tableau system for HLQL. Furthermore, we will work on possible applications of the logic, e.g., to the mobile robot’s
navigation or to assistance to a blind person.
Acknowledgments.
This research is partially supported by the Polish National Science Centre grant
2011/02/A/HS1/00395.
References
[1] M. Aiello, I. Pratt-Hartmann, J. van Benthem, What is spatial logic?, in:
M. Aiello, I. Pratt-Hartmann, J. van Benthem (eds.), Handbook of Spatial
Logics, Springer, 2007, pp. 1–11.
[2] M. Duckham, J. Lingham, K. Mason, M. Worboys, Qualitative reasoning about
consistency in geographic information, Information Sciences 176 (6) (2006) 601–
627.
38
[3] M. Eppe, M. Bhatt, Narrative based postdictive reasoning for cognitive robotics,
arXiv preprint arXiv:1306.0665.
[4] C. Lutz, F. Wolter, Modal logics of topological relations, arXiv preprint
cs/0605064.
[5] M. Marx, M. Reynolds, Undecidability of compass logic, Journal of Logic and
Computation 9 (6) (1999) 897–914.
[6] A. Montanari, G. Puppis, P. Sala, A decidable spatial logic with cone-shaped
cardinal directions, in: Computer Science Logic, Springer, 2009, pp. 394–408.
[7] A. Morales, I. Navarrete, G. Sciavicco, A new modal logic for reasoning about
space: spatial propositional neighborhood logic, Annals of Mathematics and
Artificial Intelligence 51 (1) (2007) 1–25.
[8] D. A. Randell, Z. Cui, A. G. Cohn, A spatial logic based on regions and connection., KR 92 (1992) 165–176.
[9] J. Renz, B. Nebel, Qualitative spatial reasoning using constraint calculi, in:
Handbook of spatial logics, Springer, 2007, pp. 161–215.
[10] Y. Venema, et al., Expressiveness and completeness of an interval tense logic.,
Notre Dame Journal of Formal Logic 31 (4) (1990) 529–547.
39
Algebra-based canonical formulas for
superintuitionistic logics
Julia Ilin
Superintuitionistic logics (si-logics for short) are propositional logics extending
intuitionistic propositional calculus (IPC). Finding uniform axiomatizations for
si-logics has been a significant problem in this area. Axiomatization methods via the
so-called frame-based formulas were studied by Jankov, de Jongh, Fine and later by
Zakharyaschev (see e.g. [4] for an overview). Extending the approaches of Jankov,
de Jongh and Fine, Zakharyaschev defined canonical formulas that axiomatize
all si-logics. Among other things Zakharyaschev’s canonical formulas provide a
transparent alternative proof of the Blok-Esakia theorem. In addition, canonical
formulas give rise to particular classes of si-logics. For instance, the well-known
classes of subframe and cofinal subframe logics are exactly the logics axiomatizable
by canonical formulas where a specific parameter is restricted. Zakharyaschev also
defined canonical formulas for modal logics that axiomatize all normal extensions of
K4. Jeřábek [5] generalized canonical formulas to canonical multi-conclusion rules
that axiomatize modal and intuitionistic multi-conclusion consequence relations.
Zakharyaschev’s method is model-theoretic in nature. Recently, an algebrabased approach to canonical formulas for si-logics has been developed in [1, 2]. The
algebraic perspective reveals that the main technical tool used in the theory of
canonical formulas is the disjunction-free reduct of Heyting algebras. This reduct
is locally finite1 , which is the key property for this method. This insight promises
generalizations of the method of algebra-based canonical formulas to other nonclassical logics that have algebraic semantics. In this talk we provide an overview on
this algebra-based approach to canonical formulas for si-logics and point towards
future research using this method.
The algebraic counterpart of Zakharyaschev’s canonical formulas for si-logics are
called (∧, →)-canonical formulas. The (∧, →)-canonical formulas fully describe the
disjunction-free reduct of a finite subdirectly irreducible (s.i.) Heyting algebra A, but
describe the structure of the missing connective ∨ only partially on a subset D ⊆ A2 .
The (∧, →)-canonical formulas, though syntactically quite different, serve the same
purpose as Zakharyaschev’s formulas, namely providing uniform axiomatizations of
1
Recall that a variety V is locally finite iff every finitely generated algebra A ∈ V is finite. It is
well known that the variety of Heyting algebras is not locally finite. In fact, already the free Heyting
algebra with one generator, the so-called Rieger-Nishimura lattice, is infinite.
40
all si-logics. In fact, any given axiomatization of some si-logic can be transformed
effectively into an axiomatization of the logic in terms of (∧, →)-canonical formulas.
In analogy with Zakharyaschev’s canonical formulas, (∧, →)-canonical formulas
give rise to subframe and cofinal subframe logics. These logics are axiomatized by
(∧, →)-canonical formulas where D = ∅ and depending on whether the behavior of ¬
is taken into account. Subframe and cofinal subframe logics are well-studied large
classes of si-logics with good properties such as the finite model property (FMP). In
algebraic terms, a logic is a subframe logic if its corresponding variety of Heyting
algebras is closed under (∧, →)-subalgebras and a cofinal subframe logic if it is closed
under (∧, → 0)-subalgebras.
The proof method used to establish the theory of (∧, →)-canonical formulas relies
on the locally finiteness of the disjunction-free reduct of Heyting algebras. Besides
the (∧, →)-reduct, Heyting algebras have other well-known locally finite reducts.
As a matter of fact, some of these are suitable candidates for different kinds of
algebra-based canonical formulas.
Indeed, in [2] canonical formulas based on the bounded lattice reduct of Heyting
algebras were introduced. These formulas were called (∧, ∨)-canonical formulas.
They fully describe the bounded lattice structure of a finite s.i. Heyting algebra
A and only partially the →-structure on the domain prescribed by an additional
parameter D ⊆ A2 . As in the (∧, →)-case, the (∧, ∨)-canonical formulas axiomatize
all si-logics. Neglecting the parameter D in (∧, ∨)-canonical formulas leads to stable
si-logics that form the (∧, ∨)-analogue of subframe logics. Stable logics form a
well-behaved class of logics, in particular, they have the finite model property. In
fact, many well-known si-logics are stable and in [2] it was shown that there is a
continuum of stable logics.
In this talk, we will provide an overview of the algebra-based approach to canonical
formulas. We will focus on algebra-based formulas for si-logics. In particular, we
will sketch the algebraic proof of the main result of canonical formulas stating that
all si-logics are axiomatizable by these formulas. Moreover, we will give examples
of si-logics axiomatized by particular canonical formulas. We will also present new
results on cofinal stable logics that form the (∧, ∨)-analogue to cofinal subframe
logics (see [3] for details). Finally, we will discuss how the method of algebra-based
canonical formulas can be applied to other non-classical logics. Indeed, it seems to
be a major feature of this axiomatization method that it can be applied to many
classes of non-classical propositional logics with suitable algebraic semantics. This
has been done for the case of modal logics. But the method promises to be applicable
to wider classes of logics such as substructural logics2 , or multi-modal logics such as
PDL. We will discuss ongoing work in these directions.
2
As a start, canonical formulas for the k-potent fragment of Lambek calculus have been presented
recently at LATD 2014 by N. Bezhanishvili, N. Galatos and L. Spada.
41
References
[1] G. Bezhanishvili and N. Bezhanishvili. “An algebraic approach to canonical
formulas: Intuitionistic case". Review of Symbolic Logic 2.3 (2009), pp. 517–549.
[2] G. Bezhanishvili and N. Bezhanishvili. “Locally finite reducts of Heyting algebras
and canonical formulas". To appear in Notre Dame Journal of Formal Logic.
[3] G. Bezhanishvili, N. Bezhanishvili and J.Ilin. “Cofinal stable logics". Available
at http://www.illc.uva.nl/Research/Publications/Reports/PP-2015-08.text.pdf.
[4] A. V. Chagrov and M. Zakharyaschev. Modal Logic. Vol. 35. Oxford logic guides.
Oxford University Press, 1997.
[5] E. Jeřábek. “Canonical rules". Journal of Symbolic Logic 74.4 (2009), pp. 1171–
1205.
42
Towards Reconciling SPARQL and Certain
Answers
Shqiponja Ahmetaj and Wolfgang Fischl
In the recently released recommendation [6], the W3C has defined various
SPARQL entailment regimes to allow users to specify implicit knowledge about
the vocabulary in an RDF graph. The theoretical underpinning to the systems
for query answering under rich entailment regimes is provided by the big body of
work on ontology-based query answering, notably in the area of Description Logics
(DLs) [3]. However, the semantics of query answering under SPARQL entailment
regimes is defined in a more naive and much less expressive way than the certain
answer semantics usually adopted in the DL and database literature.
Example 1. Consider an RDF graph G containing a single triple (b, a, Prof) – stating
that b is a professor – and an ontology O containing the triples
(Prof, rdfs:sc, _:b), (_:b, a, owl:Restriction),
(_:b, owl:onProperty, teaches), (_:b, owl:someValuesFrom, owl:Thing).
– stating that every professor teaches somebody. Now consider the following simple
SPARQL query: SELECT ?x WHERE (?x, teaches, ?y).1 According to the SPARQL
entailment regimes standard [6], this query yields as result the empty set.
This result is rather unintuitive: by the inclusion we know for certain that b
teaches somebody. However, the SPARQL entailment standard requires that all
values assigned to any variable must come from the RDF graph – thus treating
distinguished variables (which are ultimately output) and non-distinguished variables
(which are eventually projected out) in the same way. In contrast, the certain answer
semantics retrieves all mappings on the distinguished variables that allow to satisfy
the query in every possible model of the database and the ontology (yielding the
certain answer µ = {?x → b} in the above example).
The goal of this work is to introduce an intuitive certain answer semantics
also for SPARQL under OWL 2 QL entailment with similarly favorable results as for
CQ answering under DL-LiteR (which provides the theoretical underpinning of the
OWL 2 QL entailment regime).
The reason why for this purpose we cannot simply take over all the results from
CQ answering under DL-Lite is that SPARQL provides some crucial extensions over
1
Following [10], we use a more algebraic style notation, denoting triples in parentheses with
comma-separated components, rather than the blank-separated turtle notation.
43
CQs. One of them is the OPTIONAL operator (henceforth referred to as OPT
operator, for short). It allows the user to retrieve partial solutions in cases where no
match for the complete query can be found, instead of failing to provide any solution.
Observe that these queries are no longer monotone. Thus, the usual certain answer
semantics (i.e., something is a certain answer if it is present in every model) turns
out to be unsatisfactory:
Example 2. Consider the SPARQL query:
SELECT
?x, ?z
WHERE
(?x, teaches, ?y)
OPT
(?y, knows, ?z)
over
the
graph
G = {(b, teaches, c)} and empty ontology O. The query yields as only solution the
mapping µ = {?x → b}. Clearly, also the extended graph G0 = G ∪ {(c, knows, d)} is
a model of (G, O). But in G0 , µ is no longer a solution since µ can be extended to
µ0 = {?x → b, ?z → d}. Hence, there exists no mapping which is a solution in every
possible model of (G, O).
In this work, we investigate further problems with a literal adoption of a certain
answer semantics in the presence of the OPT operator, and propose a suitable
modified definition for the class of well-designed SPARQL queries [10]. This modified
semantics also requires an adaptation and extension of the known query answering
algorithms for DL-Lite. We present two such modified algorithms for query evaluation.
Related Work to our findings includes the work our approaches are based upon
[2, 4, 5, 6]. There is a huge body of results on CQ answering in DLs (cf. [4, 5, 9]). For
SPARQL recent work [7] presents a stronger semantics.In [1], the authors describe a
rewriting of SPARQL query answering under OWL 2 QL into Datalog± . A slight
modification allows them to remove the active domain semantics of variables, however
this only applies to variables occuring in a single BGP. Libkin [8] also criticizes
the standard notion of certain answers in case of non-monotone queries. Similar
to his suggestion to use the greatest lower bounds in terms of informativeness, our
approach chooses the most informative solutions as certain answers.
Acknowledgements
This work was supported by the Vienna Science and Technology Fund (WWTF),
project ICT12-15 and by the Austrian Science Fund (FWF): P25207-N23.
References
[1] M. Arenas, G. Gottlob, and A. Pieris. Expressive languages for querying the
semantic web. In Proc. of PODS 2014, pages 14–26. ACM, 2014.
[2] M. Arenas and J. Pérez. Querying semantic web data with SPARQL. In Proc. of
PODS 2011, pages 305–316. ACM, 2011.
44
[3] F. Baader, D. Calvanese, D. L. McGuinness, D. Nardi, and P. F. Patel-Schneider,
editors. The Description Logic Handbook: Theory, Implementation, and Applications. Cambridge University Press, 2003.
[4] D. Calvanese, G. De Giacomo, D. Lembo, M. Lenzerini, and R. Rosati. Tractable
reasoning and efficient query answering in description logics: The DL-Lite family.
J. Autom. Reasoning, 39(3):385–429, 2007.
[5] T. Eiter, M. Ortiz, M. Šimkus, T. Tran, and G. Xiao. Query rewriting for
Horn-SHIQ plus rules. In Proc. of AAAI 2012. AAAI Press, 2012.
[6] B. Glimm and C. Ogbuji. SPARQL 1.1 Entailment Regimes. W3C Recommendation, W3C, Mar. 2013. http://www.w3.org/TR/sparql11-entailment.
[7] E. V. Kostylev and B. C. Grau. On the semantics of SPARQL queries with
optional matching under entailment regimes. In Proc. of ISWC 2014, pages
374–389. Springer, 2014.
[8] L. Libkin. Incomplete data: what went wrong, and how to fix it. In Proc. PODS
2014, pages 1–13. ACM, 2014.
[9] M. Ortiz, D. Calvanese, and T. Eiter. Data complexity of query answering
in expressive description logics via tableaux. Journal of Automated Reasoning,
41(1):61–98, 2008.
[10] J. Pérez, M. Arenas, and C. Gutierrez. Semantics and complexity of SPARQL.
ACM Trans. Database Syst., 34(3), 2009.
45
Formalising the SECD machine with Nominal
Isabelle
Gergely Buday
1
Introduction
This paper describes work in progress. My doctoral research goal is to prove the
correctness of functional programs. I aim this as functional languages have formally
defined operational semantics and this makes the problem more tractable than for
imperative languages [21]. Especially this definedness makes advanced programming
concepts easier to tackle.
CharguÃľraud [6] lists three ways of verifying functional programs: first is to define
Hoare triples [9] , second is to define programs directly in a theorem prover, that
is shallow embedding and the third is to define the semantics of the language in
the logic of a theorem prover, use this definition to write programs and then prove
correctness of these, this is called deep embedding.
The SECD machine consists of four parts, the Stack, the Environment, the Control
stack and the Dump, hence the name. The stack consists of λ-expressions, used in
expression evaluation. The environment is an associative array consisting of nameclosure pairs where a closure is λ-term and an environment that stores all information
needed to evalute that term. The control stack’s elements are the remaining parts of
the expression being evaluated. The dump is either empty or contains a backup of
(S, E, C, D) triples, where D is the previous dump and this is saved when a closure is
applied from the stack. It is restored when the control stack is empty but the dump
is not.
My future plan is to prove the correctness of functional programs with respect
to an SECD-style operational semantics. For this I need a formalisation of the
SECD machine which was defined by Landin in his seminal paper [10] to evaluate
λ-expressions. This paper describes a preliminary attempt to formalise and verify
the SECD machine by contemporary tools. For a detailed introduction to the SECD
machine see [14] [7].
A formalisation states that a classical SECD machine started with a λ-expression
halts with a value that is the β-normal form of the original λ-expression.
46
2
Related work
I have found two formalisations in the literature: Ramsdell [19] and Graham [8].
Ramsdell’s verification script is available for downloading [18] as is the NQTHM-1992
theorem prover [4] [3] that it uses. The theorem prover and the verification script on
top of it ran without problems on current LISP implementations. This is of value
concerning the age of the theorem prover and of the verification script.
For modelling the α-equivalence classes of the λ-calculus, Ramsdell used De Bruijn
indices [5], which is a nameless representation of λ-expressions. That complicates
the formalisation and the handling of the numeric indices permeates other parts of
it so it would not be easy to replay the verification with other tools. The typeless
nature of LISP also adds difficulties to the formalisation.
Graham had and fulfilled a much more ambitious goal: to verify a hardware implementation of the SECD machine, down to the register level. His verification is also
available [11]. Because this formalisation is biased towards hardware implementation
its SECD model is already complex, not to mention the way down to bitly details.
So it again cannot be used directly.
Landin’s original work is quite terse on the SECD machine itself and communicates
more the idea than the details. Plotkin’s classical paper [16] has a rigorous proof
that I have chosen that to formalise. This is followed also by Field and Harrison [7],
3
Modeling the λ-calculus
To model the λ-calculus one has to equate terms that are isomorphic via a systematic
renaming of variables: clearly λx.x denotes the same function as λy.y . De Bruijn [5]
invented a nameless, numeric binder mechanism. This unfortunately involves complex
calculations on the indices and is counterintuitive regarding the way λ-expressions
are understood usually. The standard way of proofs was to use abstract syntax
and an informal argument on α-equivalence classes. There are a number of formal
approaches though: [12] [17] [1] . I have chosen the theory of nominal sets [15] and
its implementation, Nominal Isabelle [20] since it is a mature realisation and was
applied to serious examples like the π-calculus [2].
Nominal sets are built on the theory of permutation groups, one of its main concepts
is the equivariant function. Given a set X and a group G, an action of G on X is a
function G × X → X written g · x that satisfies g · (g 0 · x) = (gg 0 ) · x and e · x = x.
The full theory is described in [15].
Nominal Isabelle extends the Isabelle theorem prover [13] and the theory of nominal
sets. It is a collection of Isabelle theory files that define the notions necessary for
the theory of nominal sets, equivariant functions etc. .
It provides definitional mechanisms for nominal constructions that admit α-equivalence. Most of these lemmata can be verified by automatic methods.
47
Nominal Isabelle provides a theory of untyped λ-expressions which is based on the
following definitions:
atom-decl name
nominal-datatype lam =
Var name
| App lam lam
| Lam x::name l::lam binds x in l (Lam [-]. - [100 , 100 ] 100 )
so one can verify trivial α-equivalences easily:
lemma Lam [x]. Var x = Lam [y]. Var y
apply auto
done
4
Formalisation: a start
Plotkin [16] starts the definition of the SECD machine by a mutual definition of
closures and environments. An environment maps variable names to closures, while
a closure is a λ-term paired with an environment that contains all information to
interpret the free variables of the term. Formally:
(1) If x1 , . . . , xn are distinct variables and Cli (i = 1, n) are closures, {hxi , Cli i|i =
1, n} is in environments (n > 0).
(2) If E is an environment and M is a term such that F V (M ) ⊆ Dom(E), then
hM, Ei is a closure.
This can be translated to a nominal function definition. The current nominal package
does not support type operators such as list, so I had to encode environment lists
with the EmptyEnv ConsEnv constructor pair that imitates list definitions by Nil
and Cons, where Nil is the empty list and Cons takes an element and a list and
returns list topped with that element. The below verification is needed to ensure
that env_lookup is an equivariant function.
nominal-datatype
environment = EmptyEnv
| ConsEnv name closure environment
and closure
= Clos lam environment
nominal-function
env-lookup :: environment ⇒ name ⇒ closure
where
env-lookup EmptyEnv x = Clos (Var x) EmptyEnv
| env-lookup (ConsEnv v clos rest) x =
(if (v = x) then clos else env-lookup rest x)
apply (auto)
apply (simp add: env-lookup-graph-aux-def eqvt-def )
48
apply (rule-tac y=a in environment-closure.exhaust(1 ))
apply (auto)
done
The theorem env_lookup_graph_aux_def is the definition of the env_lookup function being defined, eqvt_def is the definition of equivariance: eqvt ?f ≡ ∀ p. p · ?f =
?f , while environment_closure.exhaust(1) is an induction theorem listing the two
clauses of the environment datatype definition.
My preliminary experience is that Plotkin’s proof is promising to use as a base for
the formalisation and Nominal Isabelle is an adequate tool for such a formalisation.
I aim to complete this formalisation.
References
[1] A. Anand and V. Rahli. A generic approach to proofs about substitution. In
Proceedings of LFMTP ’14, pages 5:1–5:8. ACM, 2014.
[2] J. Bengtson. The pi-calculus in nominal logic. Archive of Formal Proofs,
May 2012. http://afp.sf.net/entries/Pi_Calculus.shtml, Formal proof
development.
[3] R. S. Boyer and J. S. Moore. nqthm-1992. http://www.cs.utexas.edu/users/
boyer/ftp/nqthm/. Accessed: 2014-09-25.
[4] R. S. Boyer and J. S. Moore. A Computational Logic Handbook. Perspectives in
Computing. Academic Press, 1988.
[5] N. G. D. Bruijn. Lambda calculus notation with nameless dummies, a tool
for automatic formula manipulation, with application to the Church-Rosser
theorem. INDAG. MATH, 34:381–392, 1972.
[6] A. Charguéraud. Verification of call-by-value functional programs through a deep
embedding. http://www.chargueraud.org/research/2009/deep/deep.pdf,
2009.
[7] A. Field and P. Harrison. Functional Programming. International computer
science series. Addison-Wesley, 1988.
[8] B. Graham. The SECD Microprocessor: A Verification Case Study. The Springer
International Series in Engineering and Computer Science. Springer US, 1992.
[9] C. A. R. Hoare. An axiomatic basis for computer programming. Commun.
ACM, 12(10):576–580, Oct. 1969.
[10] P. J. Landin. The Mechanical Evaluation of Expressions. The Computer Journal,
6(4):308–320, Jan. 1964.
49
[11] C. Maguire. hol88 ubuntu packages.
+source/hol88. Accessed: 2014-09-25.
https://launchpad.net/ubuntu/
[12] A. Nanevski, F. Pfenning, and B. Pientka. Contextual modal type theory. ACM
Trans. Comput. Logic, 9(3):23:1–23:49, June 2008.
[13] T. Nipkow, L. C. Paulson, and M. Wenzel. Isabelle/HOL — A Proof Assistant
for Higher-Order Logic, volume 2283 of LNCS. Springer, 2002.
[14] L. C. Paulson. Foundations of functional programming. http://www.cl.cam.
ac.uk/~lp15/papers/Notes/Founds-FP.pdf, 2000.
[15] A. M. Pitts. Nominal Sets: Names and Symmetry in Computer Science, volume 57 of Cambridge Tracts in Theoretical Computer Science. Cambridge
University Press, 2013.
[16] G. D. Plotkin. Call-by-name, call-by-value and the lambda-calculus. Theor.
Comput. Sci., 1(2):125–159, 1975.
[17] N. Pouillard and F. Pottier. A unified treatment of syntax with binders. Journal
of Functional Programming, 22(4–5):614–704, Sept. 2012.
[18] J. D. Ramsdell. secd.events. http://www.ccs.neu.edu/home/ramsdell/
papers/secd.events. Accessed: 2014-09-25.
[19] J. D. Ramsdell. The tail-recursive SECD machine. J. Autom. Reasoning,
23(1):43–62, 1999.
[20] C. Urban and C. Kaliszyk. General Bindings and Alpha-Equivalence in Nominal
Isabelle. Logical Methods in Computer Science, 8(2:14), 2012.
[21] S. Winwood, G. Klein, T. Sewell, J. Andronick, D. Cock, and M. Norrish. Mind
the gap. In S. Berghofer, T. Nipkow, C. Urban, and M. Wenzel, editors, Theorem
Proving in Higher Order Logics, volume 5674 of Lecture Notes in Computer
Science, pages 500–515. Springer Berlin Heidelberg, 2009.
50
Dichotomy results for constraint satisfaction
problems
Michael Kompatscher
A constraint satisfaction problem (CSP) is a computational problem where we
are given a set of variables and a set of constraints on those variables. The task
is to decide whether there is an assignment of values to the variables that satisfies
all constraints. Computational problems of this type appear in many areas of
computer science, for example in artificial intelligence, computer algebra, scheduling,
computational linguistics, and computational biology.
Many CSPs can be modelled formally as follows: Let Γ be a structure with a
(possibly infinite) domain D and a finite relational signature. Then the constraint
satisfaction problem with template Γ, denoted by CSP(Γ), is the problem of deciding
whether a given primitive positive sentence is true in Γ.
An important question is whether for finite templates every CSP is either in the
P or NP-complete complexity class. If such a dichotomy theorem is true, then CSPs
would provide one of the largest known subsets of NP which avoids NP-intermediate
problems (under the assumption that P 6= NP).
In [1] Schaefer proved that the dichotomy result is true for templates with Boolean
domain. Special cases of his result include the NP-completeness of SAT and its two
popular variants 1-in-3 SAT and Not-All-Equal 3SAT.
In his proof Schaefer used primitive positive definable (pp-definable) relations. A
relation R is called primitive positive definable from a template Γ if R(v1 , ..., vk ) ⇔
∃x1 , ..., xm C where C is a conjunction of relations from Γ over variables {x1 , ..., xm ,
v1 , ..., vk }.
A template Γ1 is primitive positive definable from a template Γ2 , if all relations
in Γ1 are pp-definable from the relations in Γ2 . In this case CSP(Γ1 ) reduces to
CSP(Γ2 ). Using this reduction Schaefer was able to show that every CSP with a
Boolean domain that lies in P reduces to one of six known problems. All other
problems are NP-hard, because SAT reduces to them.
A modern approach uses universal algebra and polymorphisms (cf. [2]). A
polymorphism is a mapping f : Dn → D that preserves all the relations R in Γ
meaning that
(x11 , ..., x1k ) ∈ R, ..., (xn1 , ..., xnk ) ∈ R ⇒ (f (x11 , ..., xn1 ), ..., f (x1k , ..., xnk )) ∈ R.
Given a template Γ, there is a surprisingly close connection between its polymorphisms
and the computational complexity of CSP(Γ). The polymorphism clone P ol(Γ)
51
denotes the set of polymorphisms of Γ. Conversely, if O is a set of operations, then
Inv(O) denotes the set of relations having all operations in O as a polymorphism.
P ol and Inv together build a Galois connection. For any template with a finite
domain the set Inv(P ol(Γ)) is exactly the set of all relations that are pp-definable
in Γ. So instead of studying primitive positive reducts we can study the lattice of
polymorphism clones.
The universal algebraic approach even works, if we look at countably infinite,
ω-categorical templates and introduce the topology of pointwise convergence on
the clones. Then the Galois-closed objects are exactly the closed polymorphism
clones. This result has already lead to a classification of the complexity of temporal
complexity problems ([3]) and the Graph-SAT problems ([4]).
References
[1] Thomas J. Schaefer. Conference Record of the Tenth Annual ACM Symposium
on Theory of Computing ACM, New York, 1978.
[2] Hubie Chen. A rendezvous of logic, complexity, and algebra ACM Comput.
Surveys, New York, 2009.
[3] Manuel Bodirsky and Jan Kára. The complexity of temporal constraint satisfaction
problems. J. ACM 57, 2010, no. 2, Art. 9, 41 pp.
[4] Manuel Bodirsky and Michael Pinsker.
Schaefer’s theorem for graphs.
STOC’11âĂŤProceedings of the 43rd ACM Symposium on Theory of Computing,
655âĂŞ664, ACM, New York, 2011.
52
MSO model checking of graphs with
unbounded rank-width∗
Eduard Eiben
Monadic second order logic (MSO) over graphs extends first order logic by
variables that may range over sets of vertices (sometimes referred to as MSO1 logic).
For an MSO formula ϕ we can say that the MSO model-checking problem is the
problem of deciding whether a graph G satisfies a formula ϕ. Many NP-hard graph
problems can be naturally expressed as MSO model checking problems. Courcelle,
Makowsky, and Rotics in [1] showed how to solve MSO model checking efficiently
on graphs with bounded clique-width when a clique-decomposition is provided in
the input. Later, Oum and Seymour introduced the notion of rank-width [8], which
improves upon clique-width by allowing the efficient computation of decompositions
while retaining the positive algorithmic results obtained for clique-width. Rank-width
and clique-width are so far the most general parameters that allow fixed-parameter
tractable (FPT; for further details on the topic see [2]) algorithms for solving MSO
model checking.
An incomparable approach for solving graph problems is to use modulators. A
modulator of a graph G to a specified base class C is a set of vertices whose deletion
puts G in C. The cardinality of a modulator to various tractable graph classes has
long been used as a form of structure which can be exploited to obtain efficient
algorithms for a range of important problems, and various popular notions such
as vertex cover and feedback vertex set form special cases of modulators (see for
instance the work of Fellows et al. [3] or Fomin et al. [4]). In other fields of computer
science, modulators are often called backdoors and there are efficient algorithms
using small backdoors for, e.g., Satisfiability and Constraints Satisfaction
[5].
In our work, we investigate what happens when a graph contains a large but
“well-structured” (in the sense of having bounded rank-width) modulator to some
tractable class. Can such modulators still be exploited to obtain efficient algorithms?
And is it even possible to find such modulators efficiently?
We introduce a new family of parameters that capture (roughly speaking) the
minimum rank-width of any modulator to a graph class H and for any fixed H we
call our parameter simply the well-structure number or wsnH . We show that the
well-structure number is not only a more general parameter than the cardinality
of modulators, but is in fact more general than rank-width itself. Furthermore, we
∗
Joint work with Robert Ganian and Stefan Szeider.
53
provide an FPT algorithm for finding such well-structured modulators to any graph
class which can be characterized by a finite set of obstructions. This is achieved by
building on the nearly linear-time algorithm for computing split-decomposition [6]
in combination with the algorithm for computing rank-width [7].
Aside from developing algorithms utilizing well-structured modulators for individual problems, we also show that well-structured modulators can be used for the
efficient model checking of any monadic second order sentence, as long as certain
necessary conditions are met. Finally we turn our attention to the generalization of
monadic second order logic toward minimization and maximization problems. We
show that MSO optimization problems are not fixed-parameter tractable when parameterized by our parameter. This is surprising on one hand, since the fixed-parameter
tractability of MSO optimization using classical width parameters traditionally follows from the fixed-parameter tractability of MSO model checking; on the other
hand, since our parameters are strictly more general than rank-width, one cannot
expect that every problem which is FPT parameterized by rank-width would remain
FPT when parameterized by the well-structure number.
References
[1] B. Courcelle, J. A. Makowsky, and U. Rotics. Linear time solvable optimization
problems on graphs of bounded clique-width. Theory Comput. Syst., 33(2):125–
150, 2000.
[2] R. G. Downey and M. R. Fellows. Fundamentals of Parameterized Complexity.
Texts in Computer Science. Springer Verlag, 2013.
[3] M. R. Fellows, D. Lokshtanov, N. Misra, F. A. Rosamond, and S. Saurabh. Graph
layout problems parameterized by vertex cover. In ISAAC, pages 294–305, 2008.
[4] F. V. Fomin, S. Gaspers, A. V. Pyatkin, and I. Razgon. On the minimum
feedback vertex set problem: exact and enumeration algorithms. Algorithmica,
pages 293–307, 2008.
[5] S. Gaspers, N. Misra, S. Ordyniak, S. Szeider, and S. Zivny. Backdoors into
heterogeneous classes of SAT and CSP. In C. E. Brodley and P. Stone, editors,
Proceedings of the Twenty-Eighth AAAI Conference on Artificial Intelligence,
July 27 -31, 2014, Québec City, Québec, Canada., pages 2652–2658. AAAI Press,
2014.
[6] E. Gioan, C. Paul, M. Tedder, and D. Corneil. Practical and efficient split
decomposition via graph-labelled trees. Algorithmica, 69(4):789–843, 2014.
[7] P. Hliněný and S. il Oum.
Finding branch-decompositions and rankdecompositions. SIAM J. Comput., 38(3):1012–1032, 2008.
54
[8] S. Oum and P. Seymour. Approximating clique-width and branch-width. J.
Combin. Theory Ser. B, 96(4):514–528, 2006.
55
Some Cardinal Invariants of the Generalized
Baire Spaces∗
Diana Carolina Montoya Amaya
Cardinal invariants of the continuum are cardinals describing mostly the combinatorial or topological structure of the real line. They are usually defined in terms of ideals
on the reals, or some very closely related structure such us ℘(ω)/f in. Typically, they
assume values between ℵ1 , the first uncountable cardinal, and c = |2ω |= |ω ω |= |R|,
the cardinality of the continuum, so they are uninteresting under the continuum
hypothesis c = ℵ1 . However, in other models of set theory, they may assume different
values and they provide means for characterizing the structure of the real line in
such models.
The interaction between cardinal invariants in the classical Baire Space ω ω has been
deeply studied and it has contributed to the development of forcing techniques. A
new developing area of interest nowadays is to consider its natural generalizations to
the Generalized Baire Space κκ when κ is an uncountable cardinal. We will present
a summary of the existent result on this matter as well as interesting open questions.
For example, the well known Roitman’s Problem asks whether from d = ℵ1 it is
possible to prove that a = ℵ1 . Now, it is well known that d and a are independent,
i.e. it is possible to prove the consistency of both a < d and d < a. The problem that
is still open is the consistency of d = ℵ1 < a. In the uncountable case the analog of
Roitman’s Problem has already been resolved in the positive.
Theorem 1 (Theorem 2.1 in [3]). If κ is uncountable, regular cardinal and d(κ) = κ+
then a(κ) = κ+
Another example shows that, whereas in the countable case all these cardinal
invariants are at least ℵ1 , in the uncountable case the splitting number at κ may not
be at least κ+ .
Theorem 2 (Lemma 1 and 2 in [9]). s(κ) ≥ κ+ if and only if κ is weakly compact.
Finally we will show our current work in progress in the study of the cardinal
invariants in the generalization of the well known Cichoń’s Diagram. We consider
just the cardinal invariants associated to the meager ideal on κ as well as the
bounding and dominating numbers. We call this generalization Icho Diagram (See
∗
Joint work with Sy-David Friedman.
56
Fig.1). Moreover, we generalize classical models of set theory obtained as forcing
extensions (κ-Cohen, κ-Eventually Different, κ-Hechler, etc.), and study the values
that the cardinal invariants in the diagram take on them.
non M (κ)
6
b(κ)
6
κ+
- add M (κ)
- cof M (κ)
6
-
-
2κ
d(κ)
6
- cov M (κ)
Figure 1: Icho diagram (for κ strongly inaccesible)
References
[1] J. Brendle. Forcing and the Structure of the Real Line. The Bogotá Lectures.
2009
[2] T. Bartoszyński, H. Judah. On the Structure of the Real Line. A K Peters, Ltd.,
Wellesley, MA, 1995. xii+546 pp. ISBN: 1-56881-044-X MR1350295 (96k:03002)
[3] A. Blass, T. Hyttinen and Y. Zhang. Mad Families annd their Neighbors.
Preprint. 2007.
[4] J. Cummings, S. Shelah. Cardinal Invariants above the Continuum. Annals of
Pure and Applied Logic 75.(1995), no. 3, 251–268. MR1355135 (96k:03117).
[5] T. Jech. Set Theory. , second edition, Perspectives in Mathematical Logic,
Springer, Berlin, 1997. MR1492987 (99b:03061)
[6] A. Kanamori The higher infinite. Perspectives in Mathematical Logic, Springer,
Berlin, 1994. MR1321144 (96k:03125)
[7] K. Kunen. Set Theory. Studies in Logic (London), 34, College Publications,
London, 2011. MR2905394
[8] S. Shelah. On cardinal invariants of the continuum.in Axiomatic set theory (Boulder, Colo., 1983), 183–207, Contemp. Math., 31, Amer. Math. Soc., Providence,
RI. MR0763901 (86b:03064)
[9] T. Suzuki. About Splitting Numbers.Proc. Japan Acad. Ser. A Math. Sci. 74
(1998), no. 2, 33–35. MR1618475 (99m:03107)
57
Gaps In ω ω
Marlene Koelbing
We work on the Baire space ω ω , i.e. functions from N to N. At the Baire space
we use the order ≤∗ where f ≤∗ g iff f (n) ≤ g(n) for all but finitely many n ∈ N.
A (λ, µ)-pregap consists of two sequences hfα |α < λi and hgβ |β < µi such that the
fα -sequence is increasing, the gβ -sequence is decreasing and all the gβ are above all
the fα .
A pregap is a gap, if it can’t be filled. That means, there exists no h ∈ ω ω such
that fα < h for each α < λ and h < gβ for each β < µ. The first question is, for
which cardinals λ and µ (λ, µ)-gaps exist. Hadamard showed, that (ω, ω)-gaps do
not exist. That can be proved by a construction which gives an h which fills the
pregap. Hausdorff proved that an (ω1 , ω1 )-gap does exist. The construction of such
a gap works by recursion. Hausdorff and Rothberger showed the connection to the
bounding number b, which is b = min{κ : an (ω, κ) − gap exists}.
Under MartinâĂŹs Axiom the cardinality of the continuum c is also connected to the
existence of gaps. Kunen showed, that under MartinâĂŹs Axiom, if there exists a
(λ, µ)-gap for ω1 < λ ≤ µ it follows, that λ = c and additionaly, if MartinâĂŹs Axiom
holds and λ 6= ω1 is a regular cardinal less than c, then there exists a (λ, c)-gap.
The existence of (ω1 , c)-gaps and (c, c)-gaps is independent from ZFC+MartinâĂŹs
Axiom. Kunen showed, that it is possible, that both kinds of gaps exist and that
it is also consistent, that neither (ω1 , c)-, nor (c, c)-gaps exist. Yorioka proved, that
it is consistent with ZFC+MartinâĂŹs Axiom, that there exist (c, c)-gaps and no
(ω1 , c)-gap or (ω1 , c)-gaps and no (c, c)-gaps.
An other interesting question is, what happens to gaps under forcings, which do
not collapse κ and λ to ω. Do they remain gaps or can they be filled? For each
(ω1 , ω1 )-gap exists a Laver-forcing, which fills the gap and does not collapse ω1
whenever a forcing-extension exists, in which the gap can be filled. Gaps which
can be filled are called destructible. Moreover it can be proved, that the gap is
destructible, iff the correlated forcing-order has the countable chain condition. Thus
the question whether a gap is destructible can be reduced to a combinatorial question
for a partial order.
The (ω1 , ω1 )-gap, constructed by Hausdorff is not destructible. But it is consistent,
that a destructible (ω1 , ω1 )-gap exist. To show this, we can use a Cohen-forcing or
the ♦-principle.
References
[1] Teruyuki Yorioka Some results on gaps in P(ω)/fin. Kobe, 2004.
58
[2] Felix Hausdorff Die Graduierung nach dem Endverlauf. Leipzig : B. G. Teubner,
1909.
[3] Marion Scheepers Gaps in ω ω Set theory of the reals, Ramat Gan, Bar-Ilan
Univ., 1993.
59
The lifting problem for measure and category∗
Anda Ramona Tănasie
In my PhD thesis I will investigate the existence of lifting homomorphisms of the
Borel algebra modulo the ideal of meager sets on the generalized Baire space ω1ω1 .
The existence of liftings in the case of the classical Baire space (ω ω ) was already
investigated and it turned out to be independent of ZFC.
To obtain a model without liftings, Shelah defined oracle-cc iterations of forcing
notions. The purpose of my thesis is to find an appropriate generalization of this
notion and the corresponding iteration result.
Looking at the algebra of Borel sets modulo some ideal I, we have a natural
projection p : B → B/I, namely, the function mapping each Borel set to its equivalence class. We are interested whether a homomorphism h : B/I → B such that
(h(x/I ))/I = x/I exists, equivalently, a homomorphism h : B → B with Kernel I such
that h(x) = x mod I. Such a homomorphism is called lifting.
When studying the ω-case, the “reals" will be R, [0, 1], 2ω or ω ω , whichever version
is more convenient to work with in the current proof. The existence of a lifting has
been studied for the ideal Imz of Lebesgue measure zero sets and for the ideal If c of
meager sets.
Under CH, there is a lifting for B/I for both of these ideals. The proof essentially
dates back to von Neumann: for each of these ideals, given a density function
d : B → B, there is a Borel lifting h : B → B such that d(E) ⊆ h(E) for all E ∈ B
([1, Theorem 1.1]). For the existence of a density function, see [2].
A lifting homomorphism can also exist in a model of ¬CH. The construction of
a model of ZFC + ¬CH+ "there is a lifting for B/I" is due to Carlson: If ℵ2 many
Cohen reals are added to a model of CH, then in the extension c = ℵ2 and there is a
Borel lifting for B/I ( [1, Theorem 2.2]). The idea of his proof works for both Imz
and If c .
Up to this point, nothing is known about the existence of lifting homomorphisms
with c > ℵ2 .
The most impressive result concerning this problem is due to Saharon Shelah ([3],
[5] and [4]) and concludes that the existence of such a homomorphism is independent
of ZFC. He showed the consistency of ZFC + c = ℵ2 + “there is no lifting for B/Imz ".
He presented the proof for the ideal of Lebesgue measure zero sets, but the same
proof (as Shelah himself pointed out) works for the ideal of meager sets as well.
Using an ω2 stage finite support iteration of ccc forcing notions of size ℵ1 , Shelah
obtained a model V Pω2 , where there is no lifting. Obviously, in this model, CH has
to fail.
∗
Joint work with Sy-David Friedman.
60
Shelah argued that a poset killing the lifting exists, that is, a poset that adds a
Borel set X ⊆ ω ω such that h(X) cannot be defined. The oracle-cc theory comes to
use at this point for ensuring no possible value for h(X) is added in the iteration,
thus there will be no way to extend the function we had at an initial stage α to a
lifting in the final model.
References
[1] Maxim Burke. Liftings for Lebesgue measure. Set theory of the reals,(Ramat
Gan, 1991), Israel Math. Conf. Proc., volume 6 of Israel Math. Conf. Proc., pages
119âĂŞ150. Bar-Ilan Univ., Ramat Gan, 1993.
[2] Oxtoby, John C., Measure and category, volume 2 of Graduate Texts in Mathematics. Springer-Verlag, New York-Berlin, second edition, 1980. A survey of the
analogies between topological and measure spaces.
[3] Saharon Shelah. Proper forcing , volume 940 of Lecture Notes in Mathematics.
Springer-Verlag, Berlin-New York, 1982.
[4] Saharon Shelah. Lifting problem of the measure algebra , Israel J. Math.,
45(1):90âĂŞ96, 1983.
[5] Saharon Shelah. Proper and improper forcing , Perspectives in Mathematical
Logic. Springer-Verlag, Berlin, second edition, 1998.
61
Unified Correspondence as a Proof-Theoretic
Tool∗
Apostolos Tzimoulis
This presentation focuses on the formal connections between correspondence phenomena, well known from the area of modal logic, and the theory of display calculi,
originated by Belnap [1].
Sahlqvist correspondence theory. Sahlqvist theory [14, 15] is among the most
celebrated and useful results of the classical theory of modal logic, and one of the
hallmarks of its success. It provides an algorithmic, syntactic identification of a class
of modal formulas whose associated normal modal logics are strongly complete with
respect to elementary (i.e. first-order definable) classes of frames.
Unified correspondence theory. Correspondence theory is currently a very
active field of research. This field has significantly broadened its scope in recent
years, extending the benefits it originally imparted to modal logic to a plethora of
logics. These logics include, among others, intuitionistic and lattice-based (modal)
logics [7], substructural logics [8], non-normal modal logics [11, 13], hybrid logics [3],
mu-calculus [5, 4].
The breadth of this work has stimulated many and varied applications. Some are
closely related to the core concerns of the theory itself, such as the understanding
of the relationship between different methodologies for obtaining canonicity results
[12], or of the phenomenon of pseudocorrespondence [9]. Other, possibly surprising
applications include the dual characterizations of classes of finite lattices [11]. These
and other results have given rise to a theory called unified correspondence [6].
Tools of unified correspondence theory. The most important technical tools
in unified correspondence are: (a) a very general syntactic definition of the class of
Sahlqvist formulas, which applies uniformly to each logical signature and is given
purely in terms of the order-theoretic properties of the algebraic interpretations of the
logical connectives; (b) the algorithm ALBA, which effectively computes first-order
correspondents of input term-inequalities, and is guaranteed to succeed on a wide
class of inequalities (the so-called inductive inequalities) which, like the Sahlqvist
class, can be defined uniformly in each mentioned signature, and which properly and
significantly extends the Sahlqvist class.
∗
Joint work with Giuseppe Greco, Minghui Ma, Alessandra Palmigiano and Zhiguang Zhao.
62
Unified correspondence and display calculi. The proposed talk focuses on
an entirely different type of application of unified correspondence: the identification
of the syntactic shape of axioms which can be translated into analytic rules of a
display calculus. A rule is called analytic if adding it to a display calculus preserves
Belnap’s cut-elimination theorem. The connections between Sahlqvist theory and
display calculi have been seminally observed by Marcus Kracht in [10], in the context
of his characterisation of those formulas of the language of basic modal logic (which
he calls primitive formulas) which can be effectively transformed into structural rules
of display calculi.
Contributions. The two tools of unified correspondence can be put to use to
generalise Kracht’s transformation procedure from axioms into analytic rules. This
generalisation concerns more than one aspect. Firstly, in the same way in which
the definitions of Sahlqvist and inductive inequalities can be given uniformly in
each logical signature, the definition of primitive formulas/inequalities is introduced
for any logical framework the algebraic semantics of which is based on distributive
lattices with operators. Secondly, in the context of each such logical framework, we
introduce a hierarchy of subclasses of inductive inequalities, progressively extending
the primitive inequalities, the largest of which is the class of so-called analytic
inductive inequalities. This class significantly generalises the class of primitive
formulas/inequalities. We provide an effective procedure, based on ALBA, which
transforms each analytic inductive inequality into an equivalent set of analytic
rules. Moreover, we show that any analytic rule can be effectively and equivalently
transformed into some analytic inductive inequality.
Structure of the talk. We give an informal presentation of how the algorithm
ALBA computes the first-order correspondent of input inequalities. In this discussion
we highlight that the soundness of ALBA-computations is based on the order
theoretic properties of the interpretation of the logical connectives. This motivates
the identification of the syntactic shape (the so-called inductive inequalities), in any
logical signature based on distributed lattices, on which ALBA is guaranteed to be
successful. In any such signature, the theory of display calculi provides an associated
display calculus which we briefly introduce. We discuss the subclass of inductive
inequalities, the so-called analytic inductive inequalities, and we show by way of
examples how ALBA transforms these into analytic rules of the display calculus
previously introduced. Time permitting, we illustrate a different but equivalent
procedure to generate analytic rules, described in [2]. Interestingly, notwithstanding
the fact that this procedure is motivated in purely proof theoretic terms, there are
clear similarities between these two procedures. A systematic justification of these
similarities is a current research direction.
63
References
[1] Nuel Belnap. Display logic. Journal of Philosophical Logic, 11:375–417, 1982.
[2] Agata Ciabattoni and Revantha Ramanayake. Structural extensions of display
calculi: a general recipe. Proceedings of WOLLIC, 2013.
[3] W. Conradie and C. Robinson. On Sahlqvist theory for hybrid logic. Journal
of Logic and Computation, forthcoming.
[4] Willem Conradie and Andrew Craig. Canonicity results for mu-calculi: an
algorithmic approach. Journal of Logic and Computation, forthcoming.
[5] Willem Conradie, Yves Fomatati, Alessandra Palmigiano, and Sumit Sourabh.
Algorithmic correspondence for intuitionistic modal mu-calculus. Theoretical
Computer Science, 564:30–62, 2015.
[6] Willem Conradie, Silvio Ghilardi, and Alesssandra Palmigiano. Unified correspondence. In A. Baltag and S. Smets, editors, Johan F.A.K. van Benthem on
Logical and Informational Dynamics, volume 5 of Outstanding Contributions to
Logic, pages 933–975. Springer, 2014.
[7] Willem Conradie and Alessandra Palmigiano. Algorithmic Correspondence and
Canonicity for Distributive Modal Logic. Annals of Pure and Applied Logic,
163(3):338 – 376, 2012.
[8] Willem Conradie and Alessandra Palmigiano. Algorithmic correspondence
and canonicity for non-distributive logics. Journal of Logic and Computation,
forthcoming.
[9] Willem Conradie, Sam van Gool, Alessandra Palmigiano, Sumit Sourabh,
and Zhiguang Zhao. Canonicity and Relativized Canonicity via PseudoCorrespondence: an Application of ALBA. Submitted, 2014.
[10] Marcus Kracht. Power and weakness of the modal display calculus. In Proof
theory of modal logic, pages 93–121. Kluwer, 1996.
[11] Alessandra Palmigiano, Sabine Frittella, and Luigi Santocanale. Dual Characterizations for Finite Lattices via Correspondence Theory for Monotone Modal
Logic. Journal of Logic and Computation, forthcoming.
[12] Alessandra Palmigiano, Sumit Sourabh, and Zhiguang Zhao. Jónsson-style
canonicity for ALBA-inequalities. Submitted, 2014.
[13] Alessandra Palmigiano, Sumit Sourabh, and Zhiguang Zhao. Sahlqvist Theory
for Impossible Worlds. Journal of Logic and Computation, forthcoming.
64
[14] Henrik Sahlqvist. Completeness and Correspondence in the First and Second
Order Semantics for Modal Logic. In Studies in Logic and the Foundations of
Mathematics, volume 82, pages 110–143. 1975.
[15] J. van Benthem. Modal Logic and Classical Logic. Indices : Monographs in
Philosophical Logic and Formal Linguistics, Vol 3. Bibliopolis, 1985.
65
Logical Metatheorems for Abstract Spaces
axiomatized in Positive Bounded Logic
Daniel Günzel
We extend the range of the logical metatheorems due to Kohlenbach to cover
C(K), abstract Lp -spaces, Banach lattices, bands of Lp (Lq )-Bochner spaces and to
all normed structures axiomatized in positive bounded logic due to Henson-Iovino.
We also present a proof-theoretic analogon of the model-theoretic use of ultrapowers
of Banach spaces: a generalized uniform boundedness principle. This correspondence
is exemplified by several theorems having a model-theoretic and a proof-theoretic
counterpart.
1
Logical Metatheorems for Abstract Spaces axiomatized in Positive Bounded Logic
In this work my supervisor Prof. Kohlenbach and I extend the range of the logical
metatheorems developed by [5] which have found 50 applications in nonlinear analysis,
ergodic theory, partial differential equations and optimization published in the last
two decades. Our extension includes C(K), abstract Lp -spaces, Banach lattices,
bands of Lp (Lq )-Bochner spaces and all normed structures axiomatizable in positive
bounded logic.
While a “regular” theorem has mathematical objects as input such as functions,
fixed points etc., a metatheorem takes a proof of a theorem as input. More specifically,
a proof, formalized in a suitable logical framework, is used to obtain new constructive
results such as explicit quantitative bounds. Our formal framework for abstract
normed spaces X is called Aω [X, k·k], which is an extension of Peano Arithmetic
to higher types (with base types N and X). As axioms we have the axiom of
dependent choice for all types, constants and universal axioms of normed spaces
and we represent real numbers by functions f : N → N. The extensionality axiom
is not included in the theory, we only have the quantifier-free rule of extensionality
available. One of the key concepts in the proof of the metatheorem is majorizability,
going back to Howard (73), which is defined for the base type N and the abstract
type X as follows: n &N m := n ≥ m and x∗ &X x := x∗ ≥ kxk with x∗ ∈ N. For
higher types this is extended in a hereditary way. We read x∗ &ρ x as “x∗ majorizes
x”. In practice majorants are often easy to compute, for example let f : X → X
be a nonexpansive function, i.e. ∀x, y ∈ X (kf (x) − f (y)k ≤ kx − yk). Then f is
majorized by: f ∗ : N → N with f ∗ (n) := n + dkf (0)ke.
66
We discuss positive bounded logic (having the same expressive power as continuous
logic due to Keisler and adapted by [1]), which is a fragment of first-order logic due
to [3, Henson-Iovino], and is used in model theory to axiomatize classes of normed
structures which are closed under forming ultrapowers and ultraroots. Continous first
order logic is the main tool used by the above authors (see [2]), but since the expressive
power is the same we have chosen to consider the syntax of positive bounded logic.
It is a many-sorted logic, with a special sort for R. All functions are assumed to
be uniformly continuous (which is not necessary in our framework) and there are
only bounded quantifiers: ∃r xA := ∃x(kxk ≤ r ∧ A) and ∀r xA := ∀x(kxk ≤ r → A)
(with r ∈ Q positive). The prime formulas are of type t ≤ r and the connectives are
∨, ∧. A very important concept is the approximate satisfaction of a set of positive
bounded formulas. If ϕ is a positive bounded formula, ∀k ∈ Nϕ(k) expresses all 2−k
approximations of ϕ. We show that not only each positive bounded formula, which
we define as a syntactic class PBL in our framework, is admissible as an axiom in
the metatheorem, but also ∀k ∈ Nϕ(k).
Using a uniform boundedness principle (which fails in the full set-theoretic model)
of the following type (in fact we use a more technically involved axiom):
Σ01 -UBX : ∀y ρ (∀x ≤ρ y∃z ∈ NA∃ (x, y, z) → ∃z ∗ ∀x ≤ρ y∃z ≤N z ∗ A∃ (x, y, z)),
where A∃ is an ∃-formula and the type ρ may be X or N → N, we show that a
formula ϕ of the class PBL is equivalent to ∀k ∈ Nϕ(k). Now we come to one of our
main theorems, which we formulate for improved readability only for a special case:
Theorem (Logical Metatheorem). Let Θ be a set of sentences of the class PBL
and let Θ≈ be the set of all approximations of Θ. Let A∃ (f, x, T, v) be an ∃-formula
containing only f, x, T, v as free variables. Assume
Aω [X, k·k] + Θ + Σ01 -UBX ` ∀f N→N , xX , T X→X ∃v N A∃ (f, x, T, v)
then one can extract a total (bar recursive) computable functional Φ : NN ×N×NN →
N such that the following holds in all nontrivial normed spaces X satisfying the
axioms Θ≈ : for all f ∈ NN , x ∈ X, T ∈ X X and f ∗ ∈ NN , x∗ ∈ N, T ∗ ∈ NN if
f ∗ , x∗ , T ∗ & f, x, T then
∃v ≤ Φ(f ∗ , x∗ , T ∗ )A∃ (f, x, T, v).
In [5] one finds similar metatheorems for the following spaces: (complete) metric
spaces, CAT(0)-spaces, W-hyperbolic spaces, uniformly convex spaces, Hilbert spaces.
We extend this list to cover abstract Lp -spaces (1 ≤ p < ∞), Banach lattices, C(K)spaces of continuous real valued functions on an abstract compact topological space
K and bands of Lp (Lq )-Bochner spaces. This is done in the language of Banach
lattices using characterizations of Kakutani, Bohnenblust, Nakano, Gordon, Krivine,
67
and more recently for Lp (Lq ) [4, Henson-Raynaud]. For instance, by adding the
inequality
kx t ykp ≤ kxkp + kykp ≤ kx + ykp ,
for all positive x, y ∈ X
(1)
where x t y denotes the supremum of x and y (elements of a Banach lattice X), it is
known that any model of this theory (axiomatization of a Banach lattice plus (1)) is
isometrically lattice isomorphic to an Lp -space. The same technique, only changing
the norm inequality, is used to axiomatize C(K) in terms of Banach lattices. Henson
and Raynaud presented in [4] an infinite list of axioms, also using Banach lattices,
which axiomatizes bands of Lp (Lq )-Bochner spaces. In our formal framework we can
express their list of axioms by one sentence.
Finally, we provide a list of applications of the uniform boundedness principle of
the following kind: Aω [X, k·k] + Σ01 -UBX ` X is strictly convex → X is uniformly
convex. Using ultrapowers the corresponding result is, that a Banach space X is
uniformly convex if and only if its ultrapower is strictly convex. More examples of
the correspondence of uniform boundedness to the model-theoretic use of ultrapowers
are:
smoothness + Σ01 -UBX
= uniformly smoothness
X
0
normed space + Σ1 -UB
= Banach space
p(n) convexity + Σ01 -UBX = P(n)-convexity
As a summary we have established an effective and quantitative form of results
treated so far only in model theory using ultrapowers for the proof-theoretic framework. In [2] the authors have proven a completeness theorem for continuous logic,
which shall not be confused with our applied proof-theoretic approach via the above
metatheorem.
References
[1] I. Ben-Yaacov, A. Berenstein, C.W. Henson, and A. Usvyatsov, Model theory for
metric structures, Model theory with applications to algebra and analysis. vol. 2,
London Math. Soc. Lecture Note Ser., vol. 350, Cambridge Univ. Press, 2008,
pp. 315–427.
[2] I. Ben Yaacov and A.P. Pedersen, A proof of completeness for continuous firstorder logic, J. Symbolic Logic. vol. 75, 2010, pp. 168–190.
[3] C.W. Henson and J. Iovino, Ultraproducts in Analysis, In Analysis and Logic,
London Math. Soc. Lecture Note Ser., vol. 262, Cambridge Univ. Press, 2002,
pp. 1–115.
[4] C.W. Henson and Y. Raynaud, On the theory of Lp (Lq )-Banach lattices, Positivity
11 (2007), no. 2, 201–230.
68
[5] U. Kohlenbach, Applied Proof Theory: Proof Interpretations and their Use in
Mathematics, Springer Monogr. Math., Springer, Berlin, 2008.
69
Proof Mining in Nonlinear Analysis
Daniel Körnlein
Proof mining is a research program in proof theory that was inspired by Kreisel’s
central question from the 1950’s:
“What more do we know if we have proved a theorem by restricted means than if we
merely know that it is true?”
The additional content inherent in proofs in certain restricted theories can be of
quantitative (e.g. convergence rates, rates of metastability, moduli of uniqueness) or
of qualitative nature (weakened hypotheses, uniformity results). The endeavors over
the past 20 years in this vein can be divided into two categories; on the one hand,
general metatheorems guarantee the extractability of quantitative data under general
requirements on the logical form of the theorem in question and the proof-theoretic
principles used in its proof. On the other hand, there have been numerous applications
of these metatheorems to concrete examples in a variety of fields, including fixed
point theory, ergodic theory, continuous optimization, best L1 -approximation and
abstract Cauchy problems. For a comprehensive treatment see Kohlenbach [1].
The central methods used in proof mining are (variants of) Gödel’s functional
interpretation (Dialectica), negative translations and majorization. Applying (a
combination of) these methods to concrete proofs p of a theorem A results in a new
proof of p0 of a new theorem A0 , where A0 exhibits the additional content. Moreover,
A0 will imply trivially the original statement.
The talk will focus on applications of this methodology to metric fixed point theory.
In particular, we will examine from a proof-theoretic perspective a convergence result
due to Yamada [6]:
Theorem 1. Let H be a Hilbert space, T : H → H be a nonexpansive mapping
with nonempty set of fixed points F ix(T ). Suppose that for η, κ > 0, a mapping
F : H → H is κ−Lipschitzian on T (H) and satisfies the inequality hF x−F y, x−yi ≥
ηkx − yk2 for all x, y ∈ T (H). Then, for any µ and p satisfying 0 < p < 1 and
0 < µ < 2η/κ2 , the sequence (un ) defined by un := T (un−1 ) − µn−p F (T (un−1 ))
converges in norm to the unique solution of the Variational Inequality Problem VIP:
VIP: Find u∗ ∈ F ix(T ) such that hv − u∗ , F (u∗ )i ≥ 0 for all v ∈ F ix(T ).
The Cauchyness of an iteration (xn ) with respect to a metric d is translated by
the functional interpretation to
∀k ∈ N∀g : N → N∃n ∈ N∀i, j ∈ {n, n + 1, . . . n + g(n)} d(xi , xj ) < 2−k ,
70
which coincides with the Herbrand normal form (in this case!) and was recently
rediscovered by Tao under the name of metastability [4, 3]. The additional, quantitative content then consists of a bound Φ(g, k) on the existential quantifier of the
statement, called a rate of metastability.
Our analysis of this theorem will highlight the modularity of the functional
interpretation in that it is sound with respect to the modus ponens. In fact, to
obtain a rate of metastability, one can reuse an earlier analysis [2] of a theorem due
to Wittmann [5] once one has extracted the appropriate moduli from the given proof
that (an approximate version of) VIP has a solution.
References
[1] Ulrich Kohlenbach, Applied proof theory: Proof interpretations and their use in
mathematics, Springer Monographs in Mathematics, 2008.
[2]
, On quantitative versions of theorems due to F.E. Browder and R.
Wittmann, Adv. Math. 226 (2011), 2764–2795.
[3] Terence Tao, Norm convergence of multiple ergodic averages for commuting
transformations, Ergodic Theory Dynam. Systems 28 (2008), 657–688.
[4]
, Soft analysis, hard analysis, and the finite convergence principle. Essay
posted May 23, 2007. Appeared in:, T. Tao, Structure and Randomness: Pages
from Year One of a Mathematical Blog. AMS (2008), 298pp.
[5] Rainer Wittmann, Approximation of fixed points of nonexpansive mappings, Arch.
Math. (Basel) 58 (1992), 486–491.
[6] Isao Yamada, The hybrid steepest descent method for the variational inequality
problem over the intersection of fixed point sets of nonexpansive mappings, Inherently Parallel Algorithms in Feasibility and Optimization and their Applications
(Yair Censor Dan Butnariu and Simeon Reich, eds.), Studies in Computational
Mathematics, vol. 8, Elsevier, 2001, pp. 473 – 504.
71
On the Why and How of implicit conflicts in
Abstract Argumentation∗
Christof Spanring
1
Background and Motivation
This work deals with Dung-style abstract argumentation as first introduced in [4].
In a nutshell we use some structure F = (X, A) called argumentation framework
(AF), consisting of a set of arguments X and an attack relation on these arguments
A ⊆ X × X. Here arguments might refer to natural language arguments such as
a =“Homeopathy is Scam” or b =“NHS supports Homeopathy”, whereby in this
constellation naturally b attacks a but not the other way around. In graphical
representations we identify nodes with arguments and directed edges with attacks.
Example 3. Consider the AF as graphically represented in Figure 1. Here we have
that the only self-defending argument is d. If we accept d, then c is rejected, thus d
defends b. Now if we accept b then a is rejected.
NHS supports
Homeopathy
Homeopathy
is Scam
Scientific studies:
Homeopathy ⊆ Placebo
b
c
a
d
Placebo works,
and even better so
if all involved
believe it is medicine
Figure 1: Implicit Conflicts in natural language arguments.
2
Argumentation Semantics
To elaborate on a meaning of truth we use argumentation semantics (see [1] for an
established selection of such). Formally an argumentation semantics is a mapping σ,
assigning sets of reasonable arguments to each AF, σ(F) ⊆ P(X). Intuitively each
semantics builds upon a selection of semantical properties, depending only on the
∗
This research has been supported by FWF (project I1102)
72
attack relation and not on names or some underlying meaning of the arguments.
The sets S ∈ σ(F) are called σ-extensions of F. For extensions we are interested
in such properties as consistency (conflict- or attack-freeness), admissibility (selfdefense), maximality, stability (assigning to each argument exactly one status from
{accept, attack}), range-maximality (maximal area of influence), directionality (prefer
arguments that occur earlier in argumentation paths) and more.
Example 4. For Example 3 now observe that despites a and d not attacking each
other, they do not occur together in the only maximal admissible extension {b, d}.
Also observe that simply adding an attack (d, a) in this case does not change the
extensions but resolve that implicit conflict.
On the language level we can say that homeopathy might still be scam regardless
of whether placebo works or not. This objection aside, on a purely abstract level,
however, the question occurs whether implicit conflicts can always be expressed
explicitly through modification of the attack relation without altering the given
extension-sets.
3
Implicit Conflicts
Quite some effort has already been put into investigating relations between various
argumentation semantics, see e.g. [2, 6, 5]. In this work we discuss and answer questions recently posed in [3], and further investigate meaning and existence conditions
of implicit conflicts.
Definition 1. For a given AF F = (X, A) and semantics σ we say that arguments
x, y are in conflict if there is no extension S ∈ σ(F) such that x, y ∈ S. The conflict
is called explicit if (x, y) ∈ A or (y, x) ∈ A, otherwise the conflict is called implicit.
It is easy to see that for some AFs some conflicts are implicit. For (maximal)
conflict-free sets it is also easy to see that implicit conflicts occur only in relation
with self-attacking arguments. In [3] an AF was presented showing that for maximal
admissible and range-maximal admissible semantics and some AFs there is no
semantically equivalent AF with only explicit conflicts. Hence some conflicts are
implicit by their very nature and can not be explicitly represented via attacks.
Definition 2. Two AFs F and G are considered as semantically equivalent with
respect to semantics σ if they share the same sets of arguments and σ(F) = σ(G).
In this work we are interested in the mechanisms allowing implicit conflicts
and in comparisons of semantics with their way of incorporating and dealing with
(implicit) conflicts. When comparing extension-sets it occurs that more sophisticated
semantics (allowing a broader range of extension-sets) are more likely to provide
implicit conflicts, we thus regard implicit conflicts as a feature of expressiveness, and
compare various semantics with respect to this property.
73
a
a
¯
¯b
a0
b
b0
x
y
Figure 2: Example for implicit conflicts caused by directionality.
Example 5. Take the AF from Figure 2 and directionality and maximal conflictfreeness as semantical properties. As the set {x, y} has no incoming attacks, wlog.
we accept x. Then a0 is rejected and {a, a
¯} has no incoming attacks. If we accept a
then a
¯ is attacked, b0 has no more incoming attacks and can be accepted as well. But
then ¯b as well as b are rejected. Thus accepting a means that b can not be accepted
and vice versa.
If we add an attack (b, a), then {a, a
¯} receives one more incoming attack and
as second component we thus need to consider {a, a
¯, b0 , ¯b, b} instead, leading to the
¯
additional extension {x, a, b}. Observe that this AF also has an implicit conflict
between a0 and b0 which can be made explicit without altering the extension-sets.
The conflict between a and b however appears to be immanently implicit.
References
[1] Pietro Baroni, Martin Caminada, and Massimiliano Giacomin. An introduction
to argumentation semantics. Knowledge Eng. Review, 26(4):365–410, 2011.
[2] Pietro Baroni and Massimiliano Giacomin. On principle-based evaluation of
extension-based argumentation semantics. Artif. Intell., 171(10-15):675–700,
2007.
[3] Ringo Baumann, Wolfgang Dvořák, Thomas Linsbichler, Hannes Strass, and
Stefan Woltran. Compact argumentation frameworks. In Torsten Schaub, Gerhard Friedrich, and Barry O’Sullivan, editors, Proceedings of the 21st European
Conference on Artificial Intelligence (ECAI 2014), volume 263 of Frontiers in
Artificial Intelligence and Applications, pages 69–74. IOS Press, 2014.
[4] Phan Minh Dung. On the acceptability of arguments and its fundamental role in
nonmonotonic reasoning, logic programming and n-person games. Artif. Intell.,
77(2):321–357, 1995.
74
[5] Paul E. Dunne, Wolfgang Dvořák, Thomas Linsbichler, and Stefan Woltran.
Characteristics of multiple viewpoints in abstract argumentation. In Chitta
Baral, Giuseppe De Giacomo, and Thomas Eiter, editors, Proceedings of the
14th International Conference on Principles of Knowledge Representation and
Reasoning (KR 2014), pages 72–81. AAAI Press, 2014.
[6] Wolfgang Dvořák and Christof Spanring. Comparing the expressiveness of argumentation semantics. In Bart Verheij, Stefan Szeider, and Stefan Woltran,
editors, Proceedings of the 4th Conference on Computational Models of Argument (COMMA 2012), volume 245 of Frontiers in Artificial Intelligence and
Applications, pages 261–272. IOS Press, 2012.
75
Extracting consequence relations from
abstract argumentation frames
Esther Anna Corsi
In the theory on abstract argumentation introduced by Dung [1], arguments are
seen as abstract entities that may attack each other. Dung defines an argumentation
framework as a pair (Ar, →) where Ar is a set of arguments and → an attack relation
over Ar; in particular if a, b ∈ Ar, a → b means that the argument a attacks the
argument b. An argumentation framework thus is just a directed graph in which the
arguments are the vertices and the attack relation defines the edges.
Instead of considering arguments as abstract entities, we depict them as propositional formulas over the classical language (⊃, ∧, ∨ and ⊥). Often (see, e.g., [3])
an argument is identified with an ordered pair A = (Φ, φ) where Φ is a set of
propositional formulas, called the support, and φ a formula such that Φ |=CL φ,
called the claim. We will consider only the special case in which the support and
the claim coincide and these arguments may simply be identified with propositional
formulas. We say that an argumentation frame (Ar, →) is logically closed if for any
argument a ∈ Ar, all the subformulas of a are in Ar. Under this formalization, some
conditions apply to the arguments and some of them follow from what we have called
postulates. We obtain, for example, the following:
Postulate 1 (P1 ): If a |=CL b and x → b, then x → a.
From P1 , since A ∧ B |=CL A and A ∧ B |=CL B, whenever x → A (or x → B),
then x → A ∧ B. Such conditions that follow from the general ones are called weak.
Making further assumptions on the behaviour of the arguments, we define also other
conditions denominated as strong. An example of a strong condition is the following:
if x → a ∧ b, then either x → a or x → b. Similar conditions can be defined also for
the other connectives and we say that (Ar, →) is logically saturated if its arguments
satisfy these conditions.
Definition 3. Let F = (Ar, →) be a logically closed and saturated argumentation
framework, A ∈ Ar and Γ ⊆ Ar. We say that Γ |=Fatt A iff whenever x → A for some
x ∈ Ar, then there exists γ ∈ Γ such that x → γ; i.e. if x attacks the consequent
(A), then x attacks at least one element of the premise (Γ). We say that Γ |=att A iff
Γ |=Fatt A for all frames F .
For the entailment relation just introduced the following theorem holds:
Theorem 1. Γ |=att A iff Γ |=CL A, where |=CL stands for the consequence relation
for classical logic.
76
In the proof of the Theorem we use that Gentzen’s sequent calculus (see [2]) is
sound and cut-free complete with reference to classical logic and that its logical rules
are invertible.
Following Dung’s definitions [1], we say that an argument a ∈ Ar is acceptable
with respect to a set Γ ⊆ Ar of arguments iff for each argument b ∈ Ar, if b → a
then exists γ ∈ Γ such that γ → b.
Furthermore, Dung defines the characteristic function FF on an argumentation
framework as a function FF : 2Ar → 2Ar such that
FF (Γ) = {a|a is acceptable with respect to Γ}.
Therefore we can define another consequence relation |=adm whose correspondent
F
operator Cadm
coincides with FF , i.e. for every argumentation framework F logically
F (Γ) = F (Γ).
closed and saturated, and Γ ⊆ Ar, Cadm
F
Definition 4. Let F = (Ar, →) be a logically closed and saturated argumentation
framework, A ∈ Ar and Γ ⊆ Ar. We say that Γ |=Fadm A iff whenever x → a, for
some x ∈ Ar, then there exists γ ∈ Γ such that γ → x. We say that Γ |=adm A iff
Γ |=Fatt A for all frames F .
For the consequence relation just introduced the following theorem holds:
Theorem 2. Γ |=adm A iff Γ |=CL A, where |=CL stands for the consequence relation
for classical logic.
Other definitions of consequence relations similar to |=att and |=adm can be
introduced and for each of them a similar theorems to the Theorems 2 and 4 can be
proved.
These results are a starting point for a more general analysis of nonmonotonic
reasoning. A very weak form of nonmonotonic consequence can be recovered from
logical saturated argumentation frameworks as follows: A1 , . . . , An |=nm B iff no
Ai (i ∈ {1, . . . n}) attacks B. This consequence relation can be strengthened by
additionally requiring that for every argument x such that x → B, Ai → x for some
i ∈ {1, . . . , n}.
In any case the overall project proposed here is to systematically explore which
(monotonic and nonmonotonic) logic corresponds to which of the various alternative
consequence relations extracted from argumenation frameworks.
References
[1] Dung, P. M., On the acceptability of arguments and its fundamental role in
nonmonotonic reasoning, logic programming and n-person games. Artificial
Intelligence, 77: 321-357, 1995.
77
[2] Gentzen, G., Untersuchungen üdas logische Schließen, 39: 176-210, 405-431, 1935.
English translation in M. E. Szabo, editor, The Collected Papers of Gerhard
Gentzen, 68-131. North-Holland, 1969.
[3] Gorogiannis, N., Hunter, A., Instantiating abstract argumentation with classical
logic arguments: Postulates and properties, Artificial Intelligence, 175(9-10):14791497, 2011
78
The Expressive Power of k-ary
Inclusion-Exclusion Logic
Raine Rönnholm
The origin of inclusion-exclusion logic lies in the notions of dependence and
imperfect information. One of the first approaches in this area was IF-logic (independence friendly logic) by Hintikka and Sandu [9]. The truth for IF-logic was
originally defined by using semantic games of imperfect information ([10]), but an
equivalent compositional semantics was presented later by Hodges [11]. However,
in a compositional approach it is not sufficient to consider single assignments, but
instead sets of assignments which are called teams. They can be seen as parallel
positions in a semantic game, or can also be interpreted as information sets or as
databases ([14]).
By using similar team semantics, Väänänen [14] introduced dependence logic in
which the dependencies of the values of terms are interpreted on the level of atomic
formulas with dependence atoms. A similar approach is taken in independence logic
presented by Grädel and Väänänen [7]. These logics have been recently studied a lot
with an attempt to formalize the dependency phenomena in different fields of science,
such as database dependency theory ([13]), belief presentation ([3]) and quantum
mechanics ([12]).
Inclusion and exclusion logics were first presented by Galliani [4] in 2012. They
extend first order logic with inclusion and exclusion atoms similarly as dependence
atoms in dependence logic. The semantics for these atoms correspond to inclusion
and exclusion dependencies in database theory ([4]): Suppose that ~t1 , ~t2 are ktuples of terms and X is a team. The k-ary inclusion atom ~t1 ⊆ ~t2 says that the
values of ~t1 are included in the values of ~t2 in X. The k-ary exclusion atom ~t1 | ~t2
analogously says that ~t1 and ~t2 get distinct values in X. Our main topic of interest
is inclusion-exclusion logic in which we use both of these atoms that have a dualistic
relationship.
Without arity bounds exclusion logic has been shown to be equivalent with
dependence logic ([4]) which captures existential second order logic, ESO, on the
level of sentences ([14]). Inclusion logic is not comparable with dependence logic in
general ([4]), but captures positive greatest fixed point logic on the level of sentences,
as shown by Galliani and Hella [6]. Hence exclusion logic captures NP, and inclusion
logic captures PTIME over models with linear order. Galliani [4] has shown that
inclusion-exclusion logic is equivalent with independence logic, and that exactly
ESO-definable properties of teams can be defined with inclusion-exclusion logic.
79
It is natural to ask how does the arity of atoms effect the expressive power of these
logics. Hannula [8] has shown that inclusion logic has a strict arity hierarchy over
graphs, while Durand and Kontinen [2] have shown that, on the level of sentences,
k-ary dependence logic captures the fragment of ESO in which at most (k−1)-ary
functions can be quantified. Galliani, Hannula and Kontinen [5] have shown that the
latter holds also for k-ary independence logic. This however, does not resolve the
expressive power of k-ary inclusion-exclusion logic, INEX[k], since the translation
from it to independence logic does not respect the arities of atoms.
Our first theorem is that every formula of EXC[k] can be expressed with a formula
of k-ary ESO, ESO[k] (in which at most k-ary relation variables can be quantified).
The idea of this compositional translation is that for each occurrence of an exclusion
atom ~t1 | ~t2 we quantify a separate k-ary relation variable that gives limits to the
values that the tuple ~t1 can get and ~t2 cannot. We present a similar translation for
k-ary inclusion logic, INC[k], and combine these to get a translation from INEX[k]
to ESO[k].
For the other direction we show that all ESO[k]-formulas with at most k-ary free
relation variables can be expressed with a formula of INEX[k]. In this very natural
translation the quantified k-ary relation variables Pi are just replaced with k-tuples
w
~ i of quantified variables. Then we simply replace atomic formulas of the form Pi~t
~ i and formulas of the form ¬Pi~t dually with exclusion
with inclusion atoms ~t ⊆ w
atoms ~t | w
~ i.
For this latter translation we also need a new operator called term value preserving
disjunction which can be defined in INEX[k]. This is a useful operator for any logic
with team semantics, since the splitting of the team when evaluating disjunctions
tends to lose information about the values of terms. By being able to preserve values,
we can use k-tuples of first order variables to simulate k-ary relation variables as we
do in our translation from ESO[k] to INEX[k].
From these results it follows that, on the level of sentences, INEX[k] captures
the expressive power of ESO[k]. In particular, by using only unary inclusion and
exclusion atoms we get the expressive power of existential monadic second order
logic, EMSO. We also get a strict arity hierarchy for inclusion-exclusion logic on the
level of sentences, since the arity hierarchy for ESO is known to be strict as shown
by Ajtai [1] in 1983.
References
[1] M. Ajtai: Σ11 -formulae on finite structures. Annals of Pure and Applied Logic:
1–48, 1983.
[2] A. Durand and J. Kontinen: Hierarchies in Dependence Logic. ACM Trans.
Comput. Log. 13(4): 31, 2012.
80
[3] P. Galliani: The Dynamics of Imperfect Information. PhD. thesis, University of
Amsterdam, 2012.
[4] P. Galliani: Inclusion and exclusion dependencies in team semantics – On some
logics of imperfect information Annals of Pure and Applied Logic: 68–84, 2012.
[5] P. Galliani, M. Hannula and J. Kontinen: Hierarchies in independence logic. CSL
2013: 263–280.
[6] P. Galliani and L. Hella: Inclusion Logic and Fixed Point Logic. CSL 2013:
281-295.
[7] E. Grädel and J. Väänänen: Dependence and Independence. Studia Logica:
233–236, 2013.
[8] M. Hannula: Hierarchies in inclusion logic with lax semantics. ICLA 2015: 100–
118.
[9] J. Hintikka and G. Sandu: Informational Independence as a Semantical Phenomenon. Logic, Methodology and Philosophy of Science: 571–589, Amsterdam,
1989.
[10] J. Hintikka and G. Sandu: Game-Theoretical Semantics. Handbook of Logic
and Language: 361–410, Amsterdam, 1997.
[11] W. Hodges: Compositional semantics for a language of imperfect information.
Logic Journal of the IGPL, 5:539–563, 1997.
[12] T. Hyttinen, G. Paolini, J. Väänänen: Quantum Team Logic and Bell’s Inequalies. ArXiv: 1409.5537, 2015.
[13] J. Kontinen, S. Link and J. Väänänen: Independence in database relations.
WoLLIC: 179–193, 2013.
[14] J. Väänänen: Dependence Logic. Cambridge University Press, New York, 2007.
81